This paper compares the performance of three text-to-speech (TTS) models released from June 2021 to January 2022 in order to establish a baseline for Brazilian Portuguese. Those models were trained using dataset for Brazilian Portuguese. The experimental setup considers tts-portuguese dataset to fine-tune the following TTS models: VITS end-to-end model; glowtts and gradtts acoustic models both using hifi-gan vocoder. Performance metrics are arranged into objective and subjective metrics. As subjective metrics, the naturalness and intelligibility are measured based on the mean opinion score (MOS). Results shows that gradtts+hifigan model achieved naturalness of 4.07 MOS, close to performance of current commercial models.

This paper proposes two fully-automatic machine-extracted linguistic features from an unlimited text input for Mandarin prosody generation. One is the punctuation confidence (PC) which measures the likelihood of inserting a major punctuation mark (PM) at a word boundary. Another is the quotation confidence (QC) which measures the likelihood of a word string to be quoted as a meaningful or emphasized unit in text. Because a major PM in a text is highly correlated with a prosodic break, and a quoted word string plays an important role in human language understanding, the two features potentially could provide useful information for prosody generation. The idea is first realized by employing conditional random field (CRF)-based models to predict major PMs, quoted word string locations, and their associated confidences, i.e., the PC and the QC, for each word boundary. Then, the predicted punctuations and their confidences are combined with traditional contextual linguistic features to predict prosodic-acoustic features. Both objective and subjective tests showed that the prosody generation with the proposed linguistic features performed better than the one without the proposed features. So, the proposed PC and QC are promising features for Mandarin prosody generation.

The exponential increase in the use of the Internet and social media over the last two decades has changed human interaction. This has led to many positive outcomes, but at the same time it has brought risks and harms. While the volume of harmful content online, such as hate speech, is not manageable by humans, interest in the academic community to investigate automated means for hate speech detection has increased. In this study, we analyse six publicly available datasets by combining them into a single homogeneous dataset and classify them into three classes, abusive, hateful or neither. We create a baseline model and we improve model performance scores using various optimisation techniques. After attaining a competitive performance score, we create a tool which identifies and scores a page with effective metric in near-real time and uses the same as feedback to re-train our model. We prove the competitive performance of our multilingual model on two langauges, English and Hindi, leading to comparable or superior performance to most monolingual models.

Large Language Models (LLMs) have proved significant proficiency when comes to code generation especially in Structured Query Language (SQL) for databases and recent successful Text-to-SQL method involves fine-tuning pre-trained LLMs for SQL generation tasks. Transforming natural language text into SQL queries, has been attempted to solve with various learning techniques including Few-shot learning[1], fine tuning. In this paper we propose Supervised fine-tuning (SFT) as a better alternative for learning technique for text-to-SQL generation task using Code-Llama that pushes state of art accuracy on spider test suite to 89.6% on dev set which represent first instance of surpassing the earlier best-in-class with 5.5% higher score and 86.8% of exact match accuracy on dev set. Furthermore, we demonstrate that properly prompted LLM along with SFT provides far fewer hallucinations and much more robust LLM that can be used as a general tool for any text-to-SQL generation use case.

Recent advances in deep learning technology have led to increased interest in text-to-image technology, which enables computers to create images from text by simulating the human process of forming mental images. The GAN-based text-to-image technology involves the extraction of features from input text, which are combined with noise and then used as input to a GAN that generates images that are similar to the original images through competition between the generator and discriminator. Although generating images from English text is a mature area of research, text-to-image technology based on multilingualism, such as Korean, is still in its early stages of development. Webtoon is a digital comic format that allows comics to be viewed online. The creation process for webtoons is divided into story planning, content/sketching, coloring, and background drawing. Since each stage of webtoon production requires human intervention, it is both time-consuming and expensive. As a result, deep learning technologies such as automatic coloring and automatic line drawing are being used to reduce human involvement. However, there is a shortage of technology that can assist authors with story creation in webtoon production. Therefore, this study proposes a multilingual text-to-image model capable of generating webtoon images when presented with multilingual input text. The proposed model employs Multilingual BERT to extract feature vectors for multiple languages, and trains a DCGAN in conjunction with the images. The experimental results demonstrate that the model can generate images that are similar to the original images when presented with multilingual input text after training.

Text annotation is an essential element of the natural language processing approaches. The manual annotation process performed by humans has several drawbacks, such as subjectivity, slowness, fatigue, and possibly carelessness. In addition, annotators may annotate ambiguous data. So, we developed the concept of automated annotation to get the best annotations using several machine-learning approaches. The proposed approach is based on an ensemble algorithm of meta-learners and meta-vectorizer techniques. The approach employs a semi-supervised learning technique for automated annotation, aimed at detecting hate speech. This involves leveraging various machine learning algorithms, including Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbors (KNN), and Naive Bayes (NB), in conjunction with Word2Vec and TF-IDF text extraction methods. The annotation process is performed using 13,169 Indonesian YouTube comments data. The proposed model used a Stemming approach using data from Sastrawi and also new data of 2,245 words. Semi-supervised learning uses 5%, 10%, and 20% of labeled data as compared to performing labeling based on 80% of the datasets. In semi-supervised learning, the model learns from the labeled data, which provides explicit information, and the unlabeled data, which offers implicit insights. This hybrid approach enables the model to generalize and make informed predictions even when limited labeled data is available, ultimately enhancing its ability to handle real-world scenarios with scarce annotated information. In addition, the proposed method uses a variety of thresholds for matching words labeled with hate speech ranging from 0.6, 0.7, 0.8, and 0.9. The experiment showed that the KNN-Word2ec model has the best accuracy value of 96.9% with a scenario of 5%:80%:0.9. However, several other methods have also accuracy above 90%, such as SVM and DT based on both text extraction methods in several test scenarios.

In the realm of artificial intelligence (AI), generative systems, most notably Midjourney, have tremendous power to generate creative images of buildings and sites of Islamic architectural heritage through text-to-image generation based on the internet. The AI-generated representations have significant potential for architects, specialists, and common users. However, the system has considerable limitations when generating images for some buildings and sites where the representations appear too far from their original represented structures. This research article attempts to answer the question: What are the limitations of using the AI system of Midjourney in producing images similar to the original buildings and sites of the Islamic architectural heritage? The research employs prompt engineering techniques based on historical sources as inputs to examine the accuracy of the output of the AI-generated images of selected examples of structures of the Islamic tradition. It compares the output with the original look by employing direct observation and critical analysis of human intelligence (HI). It categorizes these limitations into four groups: (1) limits of the prompt, (2) limits of fame, (3) limits of regionality and historical styles, and (4) limits of architectural elements and details. It concludes that while Midjourney has great capability to represent high-end AI-generated images inspired by the Islamic tradition, it currently falls short of presenting the actual appearance of some of their original structures.

The speech-to-song illusion is a type of auditory illusion that the repetition of a part of a sentence would change people’s perception tendency from speech-like to song-like. The study aims to examine how pace, emotion, and language tonality affect people’s experience of the speech-to-song illusion. It uses a between-subject (Pace: fast, normal, vs. slow) and within-subject (Emotion: positive, negative, vs. neutral; language tonality: tonal language vs. non-tonal language) design. Sixty Hong Kong college students were randomly assigned to one of the three conditions characterized by pace. They listened to 12 audio stimuli, each with repetitions of a short excerpt, and rated their subjective perception of the presented phrase, whether it sounded like a speech or a song, on a five-point Likert-scale. Paired-sample t-tests and repeated measures ANOVAs were used to analyze the data. The findings reveal that a faster speech pace could strengthen the tendency of the speech-to-song illusion. Neither emotion nor language tonality show a statistically significant influence on the speech-to-song illusion. This study suggests that the perception of sound should be in a continuum and facilitates the understanding of song production in which speech can turn into music by having repetitive phrases and to be played in a relatively fast pace.

Transformer models have emerged as frontrunners in the field of natural language processing, primarily due to their adept use of self-attention mechanisms to grasp the semantic linkages between words in sequences. Despite their strengths, these models often face challenges in single-task learning scenarios, particularly when it comes to delivering top-notch performance and crafting strong latent feature representations. This challenge is more pronounced in the context of smaller datasets and is particularly acute for under-resourced languages such as Arabic. In light of these challenges, this study introduces a novel methodology for text classification of Arabic texts. This method harnesses the newly developed Reverse Positional Encoding (RPE) technique. It adopts inductive-transfer learning (ITL) framework combined with a switching self-attention shared encoder, thereby increasing the model's adaptability and improving its sentence representation accuracy. The integration of Mixture of Experts (MoE) and Reverse Positional Encoding (RPE) techniques empowers the model to process longer sequences more effectively. This enhancement is notably beneficial for Arabic text classification, adeptly supporting both the intricate five-point and the simpler ternary classification tasks. The empirical evidence points to its outstanding performance, achieving accuracy rates of 87.20% for the HARD dataset, 72.17% for the BRAD dataset, and 86.89% for the LABR dataset, as evidenced by the assessments conducted on these datasets.

People are drawn to woodcut-style designs due to their striking visual impact and strong contrast. However, traditional woodcut prints and previous computer-aided methods have not addressed the issues of dwindling design inspiration, lengthy production times, and complex adjustment procedures. We propose a novel network framework, the Woodcut-style Design Assistant Network (WDANet), to tackle these challenges. Notably, our research is the first to utilize diffusion models to streamline the woodcut-style design process. We've curated the Woodcut-62 dataset, featuring works from 62 renowned historical artists, to train WDANet in absorbing and learning the aesthetic nuances of woodcut prints, offering users a wealth of design references. Based on a noise reduction network, our dual cross-attention mechanism effectively integrates text and woodcut-style image features. This allows users to input or slightly modify a text description to quickly generate accurate, high-quality woodcut-style designs, saving time and offering flexibility. As confirmed by user studies, quantitative and qualitative analyses show that WDANet outperforms the current state-of-the-art in generating woodcut-style images and proves its value as a design aid.

Purpose: The present study which addressed adults who stutter (AWS), has been an attempt to investigate power spectral dynamics in stuttering state through answering the written questions using quantitative electroencephalography (qEEG).Materials and Methods: A 64-channel EEG setup was used for data acquisition in 9 AWS. Since speech, and especially stuttering, causes significant noise in the EEG, the three conditions of speech preparation (SP), imagined speech (IS), and simulated speech (SS) in a 7-band format were chosen to source localize the signals using the standard low-resolution electromagnetic tomography (sLORETA) tool in fluent and disfluent states. Results: Having extracted enough fluent and disfluent utterances, significant differences were noted. Consistent with previous studies, the lack of beta suppression in SP, especially in beta2 and beta3 and somewhat in gamma band, was localized in supplementary motor area (SMA) and pre motor area in disfluent state. Delta band frequency was the best marker of stuttering shared in all 3 experimental conditions. Decreased delta power in SMA of both hemispheres and right premotor area through SP, in fronto-central and right angular gyrus through IS, and in SMA of both hemispheres through SS were a notable qEEG features of disfluent speech. Conclusion: The dynamics of beta and delta frequency bands may potentially explain the neural networks involved in stuttering. Based on the above, neurorehabilitation may better be formulated in the treatment of speech disfluency, namely stuttering.

The article considers an approach to the problem of assessing the quality of speech during speech rehabilitation as a classification problem. For this, a classifier is built on the basis of an LSTM neural network for dividing speech signals into two classes: before the operation and immediately after. At the same time, speech before the operation is the standard to which it is necessary to approach in the process of rehabilitation. The metric of belonging of the evaluated signal to the reference class acts as an assessment of speech. An experimental assessment of rehabilitation sessions and a comparison of the resulting assessments with expert assessments of phrasal intelligibility were carried out.

Purpose: Perceptual learning and production practice are basic mechanisms that children depend on to acquire adult levels of speech accuracy. In this study, we examined perceptual learning and production practice as they contributed to changes in speech accuracy in three- and four-year-old children. Our primary focus was manipulating the order of perceptual learning and baseline production practice to better understand when and how these learning mechanisms interact. Method: Sixty-five typically-developing children between the ages of three and four were included in the study. Children were asked to produce CVCCVC nonwords like /bozjəm/ and /tʌvtʃəp/ that were described as the names of make-believe animals. All children completed two separate experimental blocks: a baseline block in which participants heard each nonword once and repeated it, and a test block in which the perceptual input frequency of each nonword varied between 1 and 10. Half of the participants completed a baseline-test order; half completed a test-baseline order. Results: Greater accuracy was observed for nonwords produced in the second experimental block, reflecting a production practice effect. Perceptual learning resulted in greater accuracy during the test for nonwords that participants heard 3 or more times. However, perceptual learning did not carry over to baseline productions in the test-baseline design, suggesting that it reflects a kind of temporary priming. Finally, a post hoc analysis suggested that the size of the production practice effect depended on the age of acquisition of the consonants that comprised the nonwords. Conclusions: The study provides new details about how perceptual learning and production practice interact with each other and with phonological aspects of the nonwords, resulting in complex effects on speech accuracy and learning of form-referent pairs. These findings may ultimately help speech-language pathologists maximize their clients’ improvement in therapy.

In software development, developers received bug reports that describe the software bug. Developers find the cause of bug through reviewing the code and reproducing the abnormal behavior that can be considered as tedious and time-consuming processes. The developers need an automated system that incorporates large domain knowledge and recommends a solution for those bugs to ease on developers rather than spending more manual efforts to fixing the bugs or waiting on Q&A websites for other users to reply to them. Stack Overflow is a popular question-answer site that is focusing on programming issues, thus we can benefit knowledge available in this rich platform. This paper, presents a survey covering the methods in the field of mining software repositories. We propose an architecture to build a recommender System using the learning to rank approach. Deep learning is used to construct a model that solve the problem of learning to rank using stack overflow data. Text mining techniques were invested to extract, evaluate and recommend the answers that have the best relevance with the solution of this bug report.

Voice cloning, an emerging field in the speech processing area, aims to generate synthetic utterances that closely resemble the voices of specific individuals. In this study, we investigate the impact of various techniques on improving the quality of voice cloning, specifically focusing on a low-quality dataset. To contrast our findings, we also use two high-quality corpora for comparative analysis. We conduct exhaustive evaluations of the quality of the gathered corpora in order to select the most suitable audios for the training of a Voice Cloning system. Following these measurements, we conduct a series of ablations by removing audios with lower SNR and higher variability in utterance speed from the corpora in order to decrease their heterogeneity. Furthermore, we introduce a novel algorithm that calculates the fraction of aligned input characters by exploiting the attention matrix of the Tacotron 2 Text-to-Speech (TTS) system. This algorithm provides a valuable metric for evaluating the alignment quality during the voice cloning process. We present the results of our experiments, demonstrating that the performed ablations significantly increase the quality of synthesised audios for the challenging low-quality corpus. Notably, our findings indicate that models trained on a 3-hour corpus from a pre-trained model exhibit comparable audio quality to models trained from scratch using significantly larger amounts of data.

The E-learning system has achieved great development after the pandemic. In this work, we proposed three artificial intelligence-based enhancements to our linguistic interactive E-learning system from different aspects. Compared with the original phonetic transcription exam system, our enhancements include an MFCC+CNN-based disordered speech classification module, a Transformer-based Grapheme-to-Phoneme converter, and a Tacotron2-based IPA-to-Speech speech synthesis system. This work not only provides a better experience for the users of this system but also explores the utilization of artificial intelligence technologies in the E-learning field and linguistic field.

The focus of this study was to investigate how individuals with bipolar disorder integrate auditory and visual speech information compared to non-disordered individuals and whether there were any differences in auditory and visual speech integration in the manic and depressive episodes in bipolar disorder patients. It was hypothesized that bipolar groups’ auditory-visual speech integration would be less robust than the control group. Further, it was predicted that those in the manic phase of bipolar disorder would integrate visual speech information more than their depressive phase counterparts. To examine these, the McGurk effect paradigm was used with typical auditory-visual speech (AV) as well as auditory-only (AO) speech perception on visual-only (VO) stimuli. Results. Results showed that the disordered and non-disordered groups did not differ on auditory-visual speech (AV) integration and auditory-only (AO) speech perception but on visual-only (VO) stimuli. The results are interpreted to pave the way for further research whereby both behavioural and physiological data are collected simultaneously. This will allow us understand the full dynamics of how, actually, the auditory and visual (relatively impoverished in bipolar disorder) speech information are integrated in people with bipolar disorder.

Hopfield Neural Networks (HNNs) are recurrent neural networks used to implement associative memory. Their main feature is their ability to pattern recognition, optimization, or image segmentation. However, sometimes it is not easy to provide the users with good explanations about the results obtained with them due to mainly the large number of changes in the state of neurons (and their weights) produced during a problem of machine learning. There are currently limited techniques to visualize, verbalize, or abstract HNNs. This paper outlines how we can construct automatic video generation systems to explain their execution. This work constitutes a novel approach to get explainable artificial intelligence systems in general and HNNs in particular building on the theory of data-to-text systems and software visualization approaches. We present a complete methodology to build these kinds of systems. Software architecture is also designed, implemented, and tested. Technical details about the implementation are also detailed and explained. Finally, we apply our approach for creating a complete explainer video about the execution of HNNs on a small recognition problem.

This paper documents the development of a special case of multilingual Automatic Speech Recognition model, specifically tailored to attend two languages spoken by the majority of Latin America, Portuguese and Spanish. The bilingual model combines Language Identification and Speech Recognition developed with the Wav2Vec2.0 architecture and trained on several open and private speech datasets. In this model, the feature encoder is trained jointly for all tasks and different context encoders are trained for each task. The model is evaluated separately on two tasks: language identification and speech recognition. The results indicate that this model achieves good performance on speech recognition and average performance on language identification, training on a low quantity of speech material. The average accuracy of the language identification module on the MLS dataset is 66.75%. The average Word Error Rate in the same scenario is 13.89%, which is better than average 22.58% achieved by the commercial speech recognizer developed by Google.

Hate Speech Detection in Arabic presents a multifaceted challenge due to the broad and diverse linguistic terrain. With its multiple dialects and rich cultural subtleties, Arabic requires particular measures to address hate speech online successfully. To address this issue, academics and developers have used natural language processing (NLP) methods and machine learning algorithms adapted to the complexities of Arabic text. However, many proposed methods were hampered by a lack of a comprehensive dataset/corpus of Arabic hate speech. In this research, we propose a novel multi-class public Arabic dataset comprised of 403,688 annotated tweets categorized as extremely positive, positive, neutral, or negative based on the presence of hate speech. Using our developed dataset, we additionally characterize the performance of multiple machine learning models for Hate speech identification in Arabic Jordanian dialect tweets. Specifically, the Word2Vec, TF-IDF, and AraBert text representation models have been applied to produce word vectors. With the help of these models, we can provide classification models with vectors representing text. After that, seven Machine learning classifiers have been evaluated: Support Vector Machine (SVM), Logistic Regression (LR), Naive Bays (NB), Random Forest (RF), AdaBoost (Ada), XGBoost (XGB), and CatBoost (CatB). In light of this, the experimental evaluation revealed that, in this challenging and unstructured setting, our gathered and annotated datasets were rather efficient and generated encouraging assessment outcomes. This will enable academics to delve further into this crucial field of study.

Speech recognition, or automatic speech recognition (ASR), is a technology designed to convert spoken language into text using software. However, conventional ASR methods involve several distinct components, including language, acoustic, and pronunciation models with dictionaries. This modular approach can be time-consuming and may influence performance. In this study, we propose a method that streamlines the speech recognition process by incorporating a unified recurrent neural network (RNN) architecture. Our architecture integrates a convolutional neural network (CNN) with an RNN and employs a connectionist temporal classification (CTC) loss function. Key experiments were carried out using a dataset comprising 576,656 valid sentences, using erosion techniques. Evaluation of the model performance, measured by the word error rate (WER) metric, demonstrated remarkable results, achieving a WER of 2%. This approach has significant implications for the realm of speech recognition, as it alleviates the need for labor-intensive dictionary creation, enhancing the efficiency and accuracy of ASR systems, and making them more applicable to real-world scenarios. For future enhancements, we recommend the inclusion of dialectal and spontaneous data in the dataset to broaden the model's adaptability. Additionally, fine-tuning the model for specific tasks can optimize its performance for targeted objectives or domains, further enhancing its effectiveness in those areas.

Nowadays, and with the mechanization of life, speech processing has become so crucial for the interaction between humans and machines. Deep neural networks require a database with enough data for training. The more features are extracted from the speech signal, the more samples are needed to train these networks. Adequate training of these networks can be ensured when there is access to sufficient and varied data in each class. If there is not enough data; it is possible to use data augmentation methods to obtain a database with enough samples. One of the obstacles to developing speech emotion recognition systems is the Data sparsity problem in each class for neural network training. The current study has focused on making a cycle generative adversarial network for data augmentation in a system for speech emotion recognition. For each of the five emotions employed, an adversarial generating network is designed to generate data that is very similar to the main data in that class, as well as differentiate the emotions of the other classes. These networks are taught in an adversarial way to produce feature vectors like each class in the space of the main feature, and then they add to the training sets existing in the database to train the classifier network. Instead of using the common cross-entropy error to train generative adversarial networks and to remove the vanishing gradient problem, Wasserstein Divergence has been used to produce high-quality artificial samples. The suggested network has been tested to be applied for speech emotion recognition using EMODB as training, testing, and evaluating sets, and the quality of artificial data evaluated using two Support Vector Machine (SVM) and Deep Neural Network (DNN) classifiers. Moreover, it has been revealed that extracting and reproducing high-level features from acoustic features, speech emotion recognition with separating five primary emotions has been done with acceptable accuracy.

Automatic Voice Query Service can extremely reduce the artificial cost, which could improve the response efficiency for users. The automatic speech recognition (ASR) is one of the important component in AVQS. However, many dialect areas in China make the AVQS have to response the multi-accented Mandarin users by single acoustic model in ASR. This problem severely limits the accuracy of ASR for multi-accented speech in the AVQS. In this paper, a new framework for AVQS is proposed to improve the accuracy of response. Firstly, the fusion feature including iVector and filterbank acoustic features is used to train the Transformer-CTC model. Secondly, the transformer-CTC model is used to construct the end-to-end ASR. Finally, key words matching algorithm for AVQS based on fuzzy mathematic theory is proposed to further improve the accuracy of response. The results show that the final accuracy in our proposed framework for AVQS arrives at 91.5%. The proposed framework for AVQS can satisfy the service requirement of different areas in mainland of China. This research has a great significance for exploring the application value of artificial intelligence in the real scene.

Electronic health records (EHRs) are becoming a vital source of data for healthcare quality improvement, research, and operations. However, much of the most valuable information contained in EHRs remains buried in unstructured text. The field of clinical text mining has advanced rapidly in recent years, transitioning from rule-based approaches to machine learning and, more recently, deep learning. With new methods come new challenges, however, especially for those new to the field. This review provides an overview of clinical text mining for those who are encountering it for the first time (e.g. physician researchers, operational analytics teams, machine learning scientists from other domains). While not a comprehensive survey, it describes the state of the art, with a particular focus on new tasks and methods developed over the past few years. It also identifies key barriers between these remarkable technical advances and the practical realities of implementation at health systems and in industry.

This study addresses the critical challenge of Speech Enhancement (SE) in noisy environments, where the deployment of Deep Neural Networksolutions on microcontrollers is hindered by their extensive computational demands. Focusing on this gap, our research introduces a novel SE method optimized for MCUs, employing a 2-layer GRU model that capitalizes on perceptual speech properties and innovative training methodologies. By incorporating self-reference signals and a dual strategy of compression and recovery based on the Mel scale, we develop an efficient model tailored for low-latency applications. Our GRU-2L-128 model demonstrates a significant reduction in size and computational requirements, achieving a 14.2× decrease in model size and a 409.1× reduction in operations compared to traditional DNN methods like DCCRN, without compromising performance. This advancement offers a promising solution for enhancing speech intelligibility in resource-constrained devices, marking a pivotal step in SE research and application.

Recently, the indications for Cochlear Implant (CI) have also been extended to subjects with Single Sided Deafness (SSD) in order to restore binaural hearing. The primary objective of this study was to investigate the advantages of CI following one year of usage in SSD subjects. Secondarily we want to assess the long-term stability of the CI. 57 participants performed a complete audiological assessment (speech audiometry in quiet and in noise and localization test). For each was also evaluated subjective hearing profile (with Speech Spatial and Qualities_SSQ questionnaire) and the impact of tinnitus (with the use of Tinnitus Handicap Inventory_THI questionnaire) After one year of CI usage, we observed a significant improvement in speech-to-noise ratio, in sound localization abilities, in the Speech (p-value 0.001) and Spatial (p-value 0.001) subscales of the SSQ questionnaire and in THI questionnaire (p-value 0.001). The improvements in speech in noise and localization abilities and the reduction of the SSQ and THI questionnaires scores were maintained over time (6th to 60th month follow-up). Only eight participants did not fully utilize the CI. CI in SSD represents an effective method for improving speech recognition in noisy environments, restoring sound localization abilities, reducing tinnitus and improve subjective listening skills.

A low-resource emotional speech synthesis system for empathetic speech synthesis based on modelling prosody features is presented here. Secondary emotions, identified to be needed for empathetic speech, are modelled and synthesised in this paper. As secondary emotions are subtle in nature, they are difficult to model compared to primary emotions. They are also less explored, and this is one of the few studies that model secondary emotions in speech. Current speech synthesis research uses large databases and deep learning techniques to develop emotion models. There are many secondary emotions, and hence, developing large databases for each of the secondary emotions is expensive. This research presents a proof-of-concept using hand-crafted feature extraction and modelling of these features using a low resource-intensive machine learning approach, thus creating synthetic speech with secondary emotions. Here, a quantitative model-based transformation is used to shape the emotional speech fundamental frequency contour. Speech rate and mean intensity are modelled via rule-based approaches. Using these models, an emotional text-to-speech synthesis system to synthesise five secondary emotions - anxious, apologetic, confident, enthusiastic and worried is developed. A perception test to evaluate the synthesised emotional speech is also conducted.

In this paper, we propose an imagined speech-based brain wave pattern recognition using deep learning. Multiple features were extracted concurrently from eight-channel Electroencephalography (EEG) signals. To obtain classifiable EEG data with fewer number of sensors, we placed the EEG sensors on carefully selected spots on the scalp. To decrease the dimensions and complexity of the EEG dataset and to avoid overfitting during the deep learning algorithm, we utilized the wavelet scattering transformation. A low-cost 8-channel EEG headset was used with MATLAB 2023a to acquire the EEG data. The Long-Short Term Memory Recurrent Neural Network (LSTM-RNN) was used to decode the identified EEG signals into four audio commands: Up, Down, Left, and Right. Wavelet scattering transformation was applied to extract the most stable features by passing the EEG dataset through a series of filtration processes. Filtration has been implemented for each individual command in the EEG datasets. The proposed imagined speech-based brain wave pattern recognition approach achieved a 92.50% overall classification accuracy. This accuracy is promising for designing a trustworthy imagined speech-based Brain-Computer Interface (BCI) future real-time systems. For better evaluation of the classification performance, other metrics were considered, and we obtained 92.74%, 92.50% and 92.62% for precision, recall, and F1-score, respectively.

Speech recognition, also known as automatic speech recognition (ASR), is a technology that enables software to transcribe spoken language into text. However, existing Amharic ASR methods require multiple separate blocks, such as language, acoustic, and pronunciation models with dictionaries, which can be time-consuming and influence performance. This study proposes an approach that replaces much of the speech pipeline with a single recurrent neural network (RNN) architecture. Our proposed architecture is based on a hybrid approach that combines a convolutional neural network (CNN) with a recurrent neural network (RNN) and a connectionist temporal classification (CTC) loss function. We conducted several experiments with noisy audio data that contain 20,000 valid sentences. The model was evaluated using the word error rate (WER) metric, achieving impressive results of 7% WER on noisy data. This approach has significant implications for the field of speech recognition, as it reduces the human effort required to create dictionaries and improves the efficiency and accuracy of ASR systems, making them more practical for real-world applications.

The large quantity of information retrieved from communities, public data repositories, web pages, or data mining can be sparsed and poorly classified. This work shows how to employ unsupervised classification algorithms such as K-means proper to classify user reviews into their closest category, forming a balanced data set. Moreover, we found that the text vectorization technique significantly impacts the clustering formation, comparing TF-IDF and Word2Vec. The value for mapping a cluster with movie genre was 81.34% ± 20.48 of the cases when the TF-IDF was applied, whereas Word2Vec only yielded a 53.51% ± 24.1. In addition, we highlight the impact of the removal of stop-words. Thus, we detected that pre-compiled lists are not the best method to remove stop-words before clustering because there is much ambiguity, centroids are poorly separated, and only 57% of clusters could match a movie genre. Thus, our proposed approach achieved a 94% of accuracy. After analyzing the classifiers’ results, we appreciated a similar effect when divided by the stop-words method removal. Statistically significant changes were observed, especially in precision metric and Jaccard scores in both classifiers, using custom-generated stop lists rather than pre-compiled ones. Reclassifying sparse data is strongly recommended as using custom-generated stop lists.

Plagiarism means taking another person’s work and not giving any credit to them for it. Plagiarism is one of the most serious problems in academia and among researchers. Even though there are multiple tools available to detect plagiarism in a document but most of them are domain-specific and designed to work in English texts, but plagiarism is not limited to a single language only. Bengali is the most widely spoken language of Bangladesh and the second most spoken language in India with 300 million native speakers and 37 million second-language speakers. Plagiarism detection requires a large corpus for comparison. Bengali Literature has a history of 1300 years. Hence most Bengali Literature books are not yet digitalized properly. As there was no such corpus present for our purpose so we have collected Bengali Literature books from the National Digital Library of India and with a comprehensive methodology extracted texts from it and constructed our corpus. Our experimental results find out average accuracy between 72.10 % - 79.89 % in text extraction using OCR. Levenshtein Distance algorithm is used for determining Plagiarism. We have built a web application for end-user and successfully tested it for Plagiarism detection in Bengali texts. In future, we aim to construct a corpus with more books for more accurate detection.

This study introduces an original comprehensive system centered on identifying specific terms that indicate a user's position, particularly the discrete values representing latitude and longitude. This system not only detects these terms but also retrieves the corresponding numerical data for accurate and efficient determination of locations. The importance of this study can be applied various fields, notably aiding offline operations of military personnel, who often lack internet access. In such scenarios, precise awareness of location is vital for strategic manoeuvres, rescue operations, and navigating unfamiliar landscapes. The system allows these personnel by allowing them to extract exact location coordinates from spoken terms, thereby enhancing their awareness even in challenging surroundings. Apart from its military utility, the project holds broader significance. Teams responding to emergencies, personnel involved in disaster management, and exploratory missions can all gain from this technology during disruptions in communication infrastructure. Furthermore, travelers, adventurers, and outdoor enthusiasts can utilize this system to accurately determine their positions in remote areas without relying on online maps. We used offline speech recognition techniques to precisely transcribe spoken terms, achieving an accuracy of over 91.3% and a word error rate of 4.2%. For sound recognition, the OpenAI Whisper model was used, and a conversion process from SpeechRecognition to AudioSegmentation was implemented, followed by transforming the audio into .wav format, we have also developed the interface of the app to use it efficiently using Streamlit. This was done to ensure seamless compatibility with the Whisper model and uninterrupted audio input. By training the system to identify specific linguistic linked to location, it achieves robust detection and extraction of relevant terms. This approach eliminates the necessity for constant internet connectivity, rendering it exceptionally useful in remote, offline, and resource-limited situations.

Detecting and understanding emotions is critical for our daily activities. As emotion recognition (ER) systems develop, we start looking at more difficult cases than just acted adult audio-visual speech. In this work, we investigate automatic classification of audio-visual emotional speech of children. Our interest is, specifically, in the improvement of the utilization of the cross-modal relationships between the selected modalities: video and audio. To underscore the importance of developing ER systems for the real-world environment, we present a corpus of children’s emotional audio-visual speech that we collected. We select a state-of-the-art model as a baseline for the purposes of comparison and present several modifications focused on a deeper learning of the cross-modal relationships. By conducting experiments with our proposed approach and the selected baseline model, we observe a relative improvement in performance by 2%. Finally, we conclude that focusing more on the cross-modal relationships may be beneficial for building ER systems for child-machine communications and the environments where qualified professionals work with children.

Purpose: We present a theoretical framework that formalizes and defines the constructs of communicative congruence and communicative dysphoria that is rooted within a comprehensive and mechanistic theory of personality. Background: Voice therapists have likely encountered a patient who states that a therapeutic target voice “isn’t me.” The ability to accurately convey a person’s sense of self, or identity, through their voice, speech, and communication behaviors seems to have high relevance to both patients and clinicians alike. However, to date, we lack a mechanistic theoretical framework through which to understand and interrogate the phenomenon of congruence between one’s communication behaviors and their sense of self. Results: We review the initial notion of congruence, first proposed by Carl Rogers. We then review several theories on selfhood, identity, and personality. After reviewing these theories, we explain how our proposed constructs fit within our chosen theory, the Cybernetic Big Five Theory of Personality. We then discuss similarities and differences to a similarly named construct, the Vocal Congruence Scale. Next, we review how these constructs may come to bear on an existing theory relevant to voice therapy, the Trans Theoretical Model of Health Behavior Change. Finally, we state testable hypotheses for future exploration, which we hope will establish a foundation for future investigations into communicative congruence. Conclusion: To our knowledge, the present paper is the first to explicitly define communicative congruence and communicative dysphoria. We embed these constructs within a comprehensive and mechanistic theory of personality and, in doing so, hope to provide a rigorous and comprehensive theoretical framework that will allow us to test and better understand these proposed constructs.

This enables the chip to regulate the voice recognition function of the gadget, leading to a more natural and human-like process. Integrating this technique into the English language voice recognition system can greatly improve its precision in identifying the speech of a particular person. This research seeks to investigate and create an automatic error detection technique for an embedded speech instruction recognition system for English using artificial intelligence. The study commences with a conventional introduction to artificial intelligence, thereafter delving into a comprehensive investigation of the speech recognition algorithm. The MATLAB programme is used to determine the accurate number of words recognised by the system and its accuracy rate. The embedded teaching recognition system for English is then tested in different contexts to assess its performance. Several tests are performed to compare and analyse the results, which indicate that the integrated English voice recognition system achieves an accuracy rate of over 90% and has a low error rate in a calm environment. In a bustling environment with numerous auditory stimuli, the accurate recognition rate typically exceeds 60%.

In this paper we describe a method which combines sentiment analysis with machine learning techniques and/or multivariate statistical analysis. By applying this methodology it is possible to classify a collection of texts into two or more groups or clusters. On the basis of a number of previously defined clusters, the novelty of the outlined approach is the use of the sentiment analysis results as input to the machine learning model or multivariate statistical analysis. Once the classifier has been obtained, we can assign a given text into one of the pre-established clusters. The groups or clusters can represent different time periods, classes of texts transcribed from different conversations, etc. The method is illustrated through an example taken from one of the two studies in which we have applied this methodology. In one of the studies, the method was used to classify press news of a volcanic eruption, while in the other study it was used to classify the conversations recorded between a chatbot with different kinds of speakers (humans or chatbots). This last study was the seminal work in which we introduced this methodology.

The digital breakthrough of the nineties has profoundly changed the way events are recognized. Starting from EST (Event Segmentation Theory) elaborated by J.M. Zacks and his collaborators, this contribution shows how the way we segment the experiential flow has been changed by the introduction of digital: while throughout modernity the perception of events was based on immediate segmentation processes, founded on five elements (agent, space, time, intention, purpose) and determined by an evident con-clusion, we currently see in the perceptive models of individuals and consequently also in aesthetic texts (films, television series, novels, graphic novels etc.) a weakening of the end. This determines not only a greater difficulty in perceiving an event, but also the need to introduce micro-caesuras between one event and another: it is space that con-stitutes the most salient category in differentiating one event from another, as is evident in the transmedia storytelling that characterizes contemporary aesthetics.

Antimicrobial resistance (AMR) is one of the top 10 threats affecting global health. AMR defeats the effective prevention and treatment of infections caused by microbial pathogens including bacteria, parasites, viruses and fungi (WHO). Microbial pathogens have natural tendency to evolve and mutate over time resulting in AMR strains. The set of genes involved in antibiotic resistance also termed as “antibiotic resistance genes” (ARGs) spread through species by lateral gene transfer thereby causing global dissemination. While this biological mechanism is prevalent in the spread of AMR, human methods also augment through various mechanisms such as over prescription, incomplete treatment, environmental waste etc. A considerable portion of scientific community is engrossed in AMR related work trying to discover novel therapeutic solutions for tackling resistant pathogens. Comprehensive inspection of the literature shows that diverse therapeutic strategies have evolved over recent years. Collectively, these therapeutic strategies include novel small molecules, newly identified antimicrobial peptides, bacteriophages, phytochemicals, nanocomposites, novel phototherapy against bacteria, fungi and virus. In this work we have developed a comprehensive knowledgebase by collecting alternative antimicrobial therapeutic strategies from literature data. We have used subjective approach for datamining new strategies resulting in broad coverage of entities and subsequently add objective data like entity name, potency, safety information etc. The extracted data was organized KOMBAT (Knowledgebase Of Microbes’ Battling Agents for Therapeutics). A lot of these data are tested against AMR pathogens. We envision that this database will be noteworthy for developing future therapeutics against resistant pathogens. The database can be accessed through http://kombat.igib.res.in/.

Bio-molecular reagents like antibodies required in experimental biology are expensive and their effectiveness, among other things, is critical to the success of the experiment. Although such resources are sometimes donated by one investigator to another through personal communication between the two, there is no previous study to our knowledge on the extent of such donations, nor a central platform that directs resource seekers to donors. In this paper, we describe, to our knowledge, a first attempt at building a web-portal titled Bio-Resource Exchange that attempts to bridge this gap between resource seekers and donors in the domain of experimental biology. Users on this portal can request for or donate antibodies, cell-lines and DNA Constructs. This resource could also serve as a crowd-sourced database of resources for experimental biology. Further, in order to index donations outside of our portal, we mined scientific articles to find instances of donations of antibodies and attempted to extract information about these donations at the finest granularity. Specifically, we extracted the name of the donor, his/her affiliation and the name of the antibody for every donation by parsing the acknowledgements sections of articles. To extract annotations at this level, we propose two approaches – a rule based algorithm and a bootstrapped relation learning algorithm. The algorithms extracted donor names, affiliations and antibody names with average accuracies of 57% and 62% respectively. We also created a dataset of 50 expert-annotated acknowledgements sections that will serve as a gold standard dataset to evaluate extraction algorithms in the future. Contact: madhavi@pitt.edu, madhavi@cs.cmu.edu Database URL: http://tonks.dbmi.pitt.edu/brx Supplementary information: Supplementary data are available at Database online.

This study combines insights from psycholinguistics and text analysis to identify linguistic markers of intercultural competence (ICC) in 1,635 blogs about intercultural experiences, written by 672 Hotel Management students. By combining holistic ICC frameworks with a text-analytical approach at word level, we were able to demonstrate that blogs with a high perceived level of ICC contain significantly more I-words, more insights words and less quantifiers. These markers of ICC constitute concrete cues for teachers when assessing reflective writing assignments and allow them to pinpoint concrete areas for improvement in their feedback and interaction with students.

Deep-learning-based speech enhancement techniques have been recently grown in interest, since their impressive performance can potentially benefit a wide variety of digital voice communication systems. However, such performance has been evaluated mostly in offline audio processing scenarios (i.e. feeding the model, in one go, a complete audio recording, which may extend several seconds). It is of great interest to evaluate and characterize the current state-of-the-art in applications that process audio online (i.e. feeding the model a sequence of segments of audio data, concatenating the results at the output end). Although evaluations and comparisons between speech enhancement techniques have been carried out before, as far as the author knows, the work presented here is the first that evaluates the performance of such techniques in relation to their online applicability. Meaning, this work measures how the output signal-to-interference ratio (as a separation metric), the response time and memory usage (as online metrics) are impacted by the input length (the size of audio segments), in addition to the amount of noise, amount and number of interferences, and amount of reverberation. Three popular models were evaluated, given their availability on public repositories and online viability: MetricGAN+, Spectral Feature Mapping with Mimic Loss, and Demucs-Denoiser. The characterization was carried out using a systematic evaluation protocol based on the Speechbrain framework. Several intuitions are presented and discussed, and some recommendations for future work are proposed.

In recent years, neural networks based on attention mechanisms have been increasingly widely used in speech recognition, separation, enhancement, and other fields. In particular, the convolution-augmented transformer has achieved good performance as it can combine the advantages of convolution and self-attention. Recently, the gated attention unit (GAU) has been proposed. Compared with the traditional multi-head self-attention, approaches with GAU are effective and computationally efficient. In this article, we propose a network for speech enhancement called CGA-MGAN, a kind of Metric GAN based on convolution-augmented gated attention. CGA-MGAN captures local and global correlations in speech signals at the same time through the fusion of convolution and gated attention units. Experiments on Voice Bank + DEMAND show that the CGA-MGAN we propose achieves an excellent performance (3.47 PESQ, 0.96 STOI, and 11.09dB SSNR) at a relatively small model size (1.14M).

Technological innovations and the expansion of Internet access have produced significant changes in the configurations of organizations and, consequently, in the relationships between employees and employers. This new scenario generates the need for greater monitoring in the workplace in order to control inappropriate behavior or situations that may generate misfortunes. Two important problems faced are the dissemination of hate through networks and data leakage that can have social, psychological, and financial impacts. Thus, monitoring tools can be incorporated to assist in surveillance, and thus ensure the achievement of organizational objectives. This paper presents a workplace computer monitoring solution that integrates Spyware techniques, and text sentiment classification, along with a distributed microservices architecture, which aims to collect a range of information and generate alerts to managers regarding hate speech and vulnerabilities. Preliminary tests have been conducted to evaluate the performance of Spyware integrated with prediction models.

Human and Computer interaction has been a part of our day-to-day life. Speech is one of the essential and comfortable ways of interacting through devices as well as a human being. The device, particularly smartphones have multiple sensors in camera and microphone, etc. speech recognition is the process of converting the acoustic signal to a smartphone as a set of words. The efficient performance of the speech recognition system highly enhances the interaction between humans and machines by making the latter more receptive to user needs. The recognized words can be applied for many applications such as Commands & Control, Data entry, and Document preparation. This research paper highlights speech recognition through ANN (Artificial Neural Network). Also, a hybrid model is proposed for audio-visual speech recognition of the Tamil and Malay language through SOM (Self-organizing map0 and MLP (Multilayer Perceptron). The Effectiveness of the different models of NN (Neural Network) utilized in speech recognition will be examined.

Conversational spoken dialogue systems that interact with the user rather than merely reading text can be equipped with hesitations to manage the dialogue flow and the users' attention. Based on a series of empirical studies, we built an elaborated hesitation synthesis strategy for dialogue systems that inserts hesitations of scalable extent wherever needed in the ongoing utterance. So far, evaluations of hesitating systems have shown that synthesis quality is affected negatively by hesitations, but that there is improvement in interaction quality. We argue that due to its conversational nature, hesitation synthesis needs interactive evaluation rather than traditional MOS-based questionnaires. To prove this point, we dually evaluate our system’s speech synthesis component: on the one hand, linked to the dialogue system evaluation, on the other hand, in the traditional MOS way. This way we are able to analyze and discuss differences that arise due to the evaluation methodology. Our results suggest that MOS scales are not sufficient to assess speech synthesis quality, which has implications for future research that are discussed in this paper. Furthermore, our results indicate that hesitations work well to increase task performance and that an elaborated strategy is necessary to avoid likability issues.

Brain-Computer Interface (BCI) systems can assist physically challenged people to interact with their surroundings and improve the quality of their lives. Decoding human thoughts is a powerful technique that can assist paralyzed people who have lost their speech production ability. Speaking is a combined process involving synchronizing the brain and the oral articulators. This paper proposed a high-accuracy brain wave pattern recognition based on inner speech using a novel feature extraction method. Only eight EEG electrodes were used in this study, and they were set on selected spots on the scalp. Support Vector Machine (SVM) was employed to decode the recorded EEG dataset into four internally spoken words which are: Up, Down, Left, and Right. The proposed approach achieved overall classification accuracy that ranges between 96.20% to 97.5%. In addition, more performance evaluation metrics were estimated to test the reliability of classifying the EEG-based inner speech data, and we obtained 97.61%, 97.50%, and 97.73% for F1-score, recall, and precision respectively. Furthermore, the Area Under Curve of the Receiver Operating Characteristic (AUC-ROC) proved the strength of the proposed approach for classifying the specified inner speech commands by achieving a macro-average amount of 99.32%. The inner speech classification method using electroencephalography proposed in this work can clinically help improve communication for patients with problems including speech disorder, mutism, cognitive development, executive function, and psychopathology.

Due to the substantial growth of internet users and its spontaneous access via electronic devices, the amount of electronic contents is growing enormously in recent years through instant messaging, social networking posts, blogs, online portals, and other digital platforms. Unfortunately, the misapplication of technologies has boosted with this rapid growth of online content which leads to the rise in suspicious activities. People misuse the web media to disseminate malicious activity, perform the illegal movement, abuse other people, and publicize suspicious contents on the web. The suspicious contents usually available in the form of text, audio or video, whereas text contents have been used in most of the cases to perform suspicious activities. Thus, one of the most challenging issues for NLP researchers is to develop a system that can identify suspicious text efficiently from the specific contents. In this paper, a Machine Learning (ML)-based classification model is proposed (hereafter called STD) to classify Bengali text into non-suspicious and suspicious categories based on its original contents. A set of ML classifiers with various features has been used on our developed corpus, consisting of 7000 Bengali text documents where 5600 documents used for training and 1400 documents used for testing. The performance of the proposed system is compared with the human baseline and existing ML techniques. The SGD classifier `tf-idf’ with the combination of unigram and bigram features are used to achieve the highest accuracy of 84.57%.

American Sign Language is a popular language for deaf individuals. Communication is made easier for these people through sign language. However, in a digital era like today, there is a need for these people to be able to communicate online, and even get help from technology to communicate in person with non sign language speakers. This research will present a program able to translate American sign language to plain English. This study aims to use the OpenCV library to recognize hand signals, also a trained model to identify images so that the program can then translate them to words and letters. The program uses a data set of over 2000 images which will be in this case the largest data set available. With over 90\% of accuracy it results in a basic computer program with the largest data set available that would make possible for users to communicate with a wide variety of words and expressions.

Simple measures often couldn’t count a deep complexity. In the case of semantic complexity, conventional readability formulas share a common style, a common sort of achievements and a common borders of limitation: These formulas lack a semantics-aware approach and as a result, a precise measurement of semantic complexity couldn’t be done. In this paper, we introduce DASTEX, a novel semantics-aware complexity measure for semantic complexity of text. By DASTEX, a new layer of complexity analysis are opened for NLP, cognitive and computational tasks. This measure benefits from an intuitionistic underlying formal model which consider semantic as a lattice of intuitions. This yields to a well-defined definition for semantic of a text and its complexity. DASTEX is a practical analysis method upon this formal model. So a complete suite of idea, model and method are prepared to result in a simple but yet deep measure for semantic complexity of text. The evaluation of the proposed approach is done by 4 Experiments. The results show DASTEX is capable of measuring the semantic complexity of text in 6 application-tasks.

Electronic invoicing has become mandatory for Italian companies since January 2019. Invoices are structured in a predefined xml template where the information reported can be easily extracted and analyzed. The main aim of this paper is to exploit the information structured in electronic invoices to build an intelligent system which can facilitate accountants work. More precisely, this contribution shows how it is possible to automate part of the accounting process: all sent or received invoices of a company are classified into specific codes which represent the economic nature of the the financial transactions. In order to classify data contained in the invoices a machine learning multiclass classification problem is proposed using as input variables the information of the invoices to predict two different target variables, account codes and the VAT codes, which composes a general ledger entry. Different approaches are compared in terms of prediction accuracy. The best performance is achieved considering the hierarchical structure of the account codes.

Ontology is a formalized representation of the problem area (PrA). Representation of the PrA in the form of an domain ontology is often used in the process of development of intelligent software systems and used as a knowledge base. The process of building an ontology is complex and requires an expert in the PrA. A large number of researchers are working to solve this problem. The basis of our approach is the use of a pipeline of different linguistic methods of text analysis. The set of rules developed by us is used to build an ontology based on the content analysis of a text resource. This article describes the method of building a domain ontology based on the linguistic analysis of content of text resources, presents an example of the proposed approach, and also presents the architecture of our pipeline.

This paper presents an approach to enhancing the clarity and intelligibility of speech in digital communications compromised by various background noises. Utilizing deep learning techniques, specifically a Variational Autoencoder (VAE) with 2D convolutional filters, we aim to suppress background noise in audio signals. Our method focuses on four simulated environmental noise scenarios: storms, wind, traffic, and aircraft. Training dataset has been obtained from public sources (TED-LIUM 3 dataset, which includes audio recordings from the popular TED-TALK series) combining with these background noises. The audio signals were transformed into 2D power spectrograms, upon which our VAE model was trained to filter out the noise and reconstruct clean audio. Our results demonstrate that the model outperforms existing state-of-the-art solutions in noise sup-pression. Although differences in noise types were observed, it was challenging to definitively conclude which background noise most adversely affects speech quality. Results have been assessed with objective methods (mathematical metrics) and subjective (listening to a set of audios by humans). Notably, wind noise showed the smallest deviation between the noisy and cleaned audio, perceived subjectively as the most improved scenario. Future work involves refining the phase calculation of the cleaned audio and creating a more balanced dataset to minimize differences in audio quality across scenarios. Additionally, practical ap-plications of the model in real-time streaming audio are envisaged. This research contributes significantly to the field of audio signal processing by offering a deep learning solution tailored to various noise conditions, enhancing digital communication quality.

(1) The incidence of ankyloglossia ranges from 0.02 to 10.7%. The literature describes the effect of ankyloglossia on selected dysfunctions of the stomatognathic system, however no studies could be found reporting the influence of ankyloglossia on the occurrence of several disorders in a group of subjects. The aim of the present study was to assess the effect of lingual frenulum on swallowing, speech, occlusion, and periodontal status; (2) Methods: The subjects were 172 patients, 86 with ankyloglossia (study group) and 86 with normal tongue frenulum (control group). In all subjects, the length of tongue frenulum, the type of swallowing, tongue mobility, occlusion, periodontal status and speech abnormalities were assessed; (3) Results: All subjects from the control group and all those with mild ankyloglossia showed normal tongue mobility. A limited tongue mobility was found in 29.4% subject with moderate and in 70.6% subjects with severe ankyloglossia. Rhotacism was observed in 21.3% subjects with normal frenulum, in 2.1% with mild, 38.3% with moderate, and 38.3% with severe ankyloglossia. Malocclusion or crowding was diagnosed in subjects with mild, moderate and severe ankyloglossia in 7.4%, 33.9% and 20.7% subjects (total 62%), respectively, whereas in the control group - in 21.6% subjects. No abnormalities in the periodontium in the area of the lingual surfaces of the crowns of the lower central incisors were found in any of the examined persons. Among patients with infantile type of swallowing 24.4% had a normal length of the tongue frenulum, 11.1% - mild, 28.9% - moderate, and 35.6 - severe ankyloglossia. Among patients presenting a mature type of swallowing 58.7% had a normal length of the frenulum; (4) Conclusions: 1.A shortened tongue frenulum correlates with “infantile swallowing pattern”. 2. Moderate or severe ankyloglossia significantly limits tongue mobility. 3. Short tongue frenulum is related to speech disorders.

The information society has transformed human life. Technology is almost everywhere, including health and education. For example, years ago, speech and language therapy students required a long time and high-cost equipment to develop healthcare of the auditory and vestibular systems competencies. The high cost of the equipment permitted its practical use only in classes, hindering students’ autonomy in developing those competencies. That situation was a real issue, even more in times of pandemic where online education was essential. This article describes SAEF, an open-source software simulator for autonomously developing procedural audiology therapy competencies, user acceptance, and the validity of experiments and results. SAEF delivers immediate feedback and performance results. Obtained results permit validating students’ and educators’ acceptability of SAEF in audiology therapy education. Obtained results invite authors to continue developing simulator software solutions in other health education contexts. SAEF was developed using open-source technology to facilitate its accessibility, classification, and sustainability.

Background: In order to achieve the best possible hearing and understanding with a cochlear implant (CI), regular hearing speech therapy treatment is necessary after implantation. This treatment should also be accessible to the growing proportion of hearing-impaired people with a migration background. This requires an alternative to therapy in the therapist's native language. The aim of this study was to evaluate six multilingual conversation applications with regard to their usefulness for therapy. Material and Methods: The six most commonly used applications were reviewed in terms of accuracy of content and grammatical translation, as well as pronunciation for English, Spanish, Arabic, Turkish, and Russian by native speakers. The number of available languages, availability, cost, and additional features were also analyzed. The accuracy of the content and grammatical translation as well as the pronunciation were statistically evaluated, and the differences were highlighted. The results of the different applications were compared with the performance of a native speaker. Results: All applications tested differed significantly from the native speaker level, with Google Translator showing the closest approximation to the native speaker level. All apps offer translations for multiple languages and, with exceptions, are available in both app stores. Furthermore, all apps have additional therapist-facilitating features. Conclusion: Multilingual conversation apps can make speech therapy in a foreign language much easier when used with patients. An adaptation of the software to the specific requirements of a hearing speech therapy is necessary to achieve a linguistic level that corresponds to the native language of the therapist and to enable an easy use in the therapy.

Hateful and abusive speech presents a major challenge for all online social media platforms. Recent advances in Natural Language Processing and Natural Language Understanding allow more accurate detection of hate speech in textual streams. This study presents a multimodal approach to hate speech detection by combining Computer Vision and Natural Language processing models for abusive context detection. Our study focuses on Twitter messages and, more specifically, on hateful, xenophobic and racist speech in Greek aimed at refugees and migrants. In our approach we combine transfer learning and fine-tuning of Bidirectional Encoder Representations from Transformers (BERT) and Residual Neural Networks (Resnet). Our contribution includes the development of a new dataset for hate speech classification, consisting of tweet ids, along with the code to obtain their visual appearance, as they would have been rendered in a web browser. We have also released a pre-trained Language Model trained on Greek tweets, which has been used in our experiments. We report a consistently high level of accuracy (accuracy score=0.970, f1-score=0.947 in our best model) in racist and xenophobic speech detection.

Today’s youth have almost universal access to the internet and frequently engage in social networking activities using various social media platforms and devices. This is a phenomenon that hate groups are exploiting when disseminating their propaganda. This study seeks to better understand youth exposure to hateful material in the online space by exploring predictors of such exposure including demographic characteristics (age, gender and race), academic performance, online behaviours, online disinhibition, risk perception, and parents/guardians’ supervision of online activities. We implemented a cross-sectional study design, using a paper questionnaire, in two high schools in Massachusetts (USA), focusing on students 14 to 19 years old. Logistic regression models were used to study the association between independent variables (demographics, online behaviours, risk perception, parental supervision) and exposure to hate online. Results revealed an association between exposure to hate messages in the online space and time spent online, academic performance, communicating with a stranger on social media, and benign online disinhibition. In our sample, benign online disinhibition was also associated with students’ risk of encountering someone online that tried to convince them of racist views. This study represents an important first step in understanding youth’s risk factors of exposure to hateful material online.

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) has become an effective and widely used tool in the treatment of Parkinson’s disease (PD). STN-DBS has varied effects on speech. Clinical speech ratings suggest worsening following STN-DBS, but quantitative intelligibility, perceptual, and acoustic studies have produced mixed and inconsistent results. Improvements in phonation and declines in articulation have frequently been reported during different speech tasks under different stimulation conditions. Questions remain about preferred STN-DBS stimulation settings. Seven right-handed, native speakers of English with PD treated with bilateral STN-DBS were studied off medication at three stimulation conditions: stimulators off, 60 Hz (low frequency stimulation - LFS), and the typical clinical setting of 185 Hz (High frequency - HFS). Spontaneous speech was recorded in each condition and excerpts were prepared for transcription (intelligibility) and difficulty judgements. Separate excerpts were prepared for listeners to rate abnormalities in voice, articulation, fluency, and rate. Intelligibility for spontaneous speech was reduced at both HFS and LFS when compared to STN-DBS off. Speech produced at HFS was more intelligible than that produced at LFS, but HFS made the intelligibility task (transcription) subjectively more difficult. Both voice quality and articulation were judged to be more abnormal with STN-DBS on. STN-DBS reduced the intelligibility of spontaneous speech at both LFS and HFS but lowering the frequency did not improve intelligibility. Voice quality ratings with STN-DBS were correlated with the ratings made without stimulation. This was not true for articulation ratings. STN-DBS exacerbated an existing voice disorder and may have introduced new articulatory abnormalities.

Speech signals are degraded in real-life environments, product of background noise or other factors. The processing of such signals for voice recognition and voice analysis systems presents important challenges. One of the conditions that make adverse quality difficult to handle in those systems is reverberation, produced by sound wave reflections that travel from the source to the microphone in multiple directions.To enhance signals in such adverse conditions, several deep learning-based methods have been proposed and proven to be effective. Recently, recurrent neural networks, especially those with long and short-term memory (LSTM), have presented surprising results in tasks related to time-dependent processing of signals, such as speech. One of the most challenging aspects of LSTM networks is the high computational cost of the training procedure, which has limited extended experimentation in several cases. In this work, we present a proposal to evaluate the hybrid models of neural networks to learn different reverberation conditions without any previous information. The results show that some combination of LSTM and perceptron layers produce good results in comparison to those from pure LSTM networks, given a fixed number of layers. The evaluation has been made based on quality measurements of the signal's spectrum, training time of the networks and statistical validation of results. Results help to affirm the fact that hybrid networks represent an important solution for speech signal enhancement, with advantages in efficiency, but without a significan drop in quality.

We present an Arduino-based automatic robotic system which is used for cutting grass or lawns, mostly healthy grass which needs to cut neatly like in a public park or a private garden. The purpose of this proposed project is to design a programmable automatic pattern design grass cutting robot with solar power which no longer requires time-consuming manual grass-cutting, and that can be operated wirelessly using an Android Smartphone via Bluetooth from a safe distance which is capable of cutting the grass in indeed required shapes and patterns; the cutting blade can also be adjusted to maintain the different length of the grass. The main focus was to design a prototype that can work with a little or no Physical user interaction. The proposed work is accomplished by using an Arduino microcontroller, DC geared Motors, IR obstacle detection sensor, motor shield, relay module, DC battery, solar panel, and Bluetooth module. The grass-cutting robot system can be moved to the location in the lawn remotely where the user wants to cut the grass directly or in desired patterns. The user can press the desired pattern button from the mobile application, and the system will start cutting grass in the similar design such as a circle, spiral, rectangle, and continue pattern. Also, with the assistance of sensors positioned at the front of the vehicle, an automatic barrier detection system is introduced to enhance safety measurements to prevent any risks. IR obstacle detector sensors are used to detect obstacles, if any obstacle is found in front of the robot while traveling; it avoids the barrier by taking a right/right turn or stop automatically appropriately, thereby preventing the collision. Also, the main aim of this project is the formation of a grass cutter that relieves the user from mowing their own grasses and reduces environmental and noise pollution. The proposed system is designed as a lab-scale prototype to experimentally validate the efficiency, accuracy, and affordability of the systems. The experimental results prove that the proposed work has all in one capability (Simple and Pattern based grass cutting with mobile-application, obstacle detection), is very easy to use, and can be easily assembled in a simple hardware circuit. We note that the systems proposed can be implemented on a large scale under real conditions in the future, which will be useful in robotics applications and cutting grass in playing grounds such as cricket, football, and hockey, etc.

This study investigates the effects of ambient illumination and negatively polarized text color on visual fatigue, exploring the issue of visual fatigue when using visual display terminals in low-illumination environments. The research methodology utilized an experimental design to collect data on changes in pupil accommodation and blink rate through an eye tracker. Partici-pants completed a reading task while exposed to various text colors and ambient light conditions to evaluate visual fatigue and cognitive performance. The study's findings suggest that text color significantly affects visual fatigue, with red text causing the highest level of visual fatigue and yellow text causing the lowest level of visual fatigue. Improvements in ambient lighting reduce visual fatigue, but the degree of improvement varies depending on the text color. Additionally, cognitive performance is better when using yellow and white text but worse when using red text. Yellow text is the most effective choice for reducing visual fatigue under negative polarity. In-creasing ambient lighting can also improve visual fatigue in low-illumination conditions. These findings will offer valuable guidance for designing visual terminal device interfaces, especially for low-illumination or night environments, to minimize visual fatigue and improve the user ex-perience.

With the abundance of conversations happening everywhere, dialogue summarization plays an increasingly important role in the real world. However, dialogues inevitably involve lots of personal pronouns, which hinder the performance of existing dialogue summarization models. This work proposes a framework named WHORU to inject external personal pronoun resolution (PPR) information into abstractive dialogue summarization models. To reduce time and space consumption, we further propose a simple and effective PPR method for the dialogue domain. Experiments demonstrated the superiority of the proposed methods. More importantly, we achieve new SOTA results on SAMSum and AMI datasets.

This research used an experimental study design in which 60 subjects were randomly assigned to either an experimental (image e-book guidance) group or a control (text guidance) group after meeting the admission conditions and agreeing to participate in the study. It was proven that providing image e-book guidance before surgery significantly changed the behavior of patients and promoted relief of urinary catheter discomfort through self-efficacy to reduce urinary catheter pain after surgery. It was found that providing image e-book guidance can shorten the time for health education and provide patients with repeated medical education and familiarity with health guidance, which can help to address the important clinical service demand issue and the shortage of nursing staff.

As RPG has high sales and profits, lots of developers have supplied various RPG to market but it changed to mass production type with sensational advertising, low quality and excessive charging and similar contents which affects game market and users’ game play experience. The author of this paper studied ways to improve mobile RPG by collecting and analyzing users’ reviews using crawling on Google Play Store. The author of this paper used topic modeling that uses text mining technique and LDA (Latent Dirichlet Allocation) to extract meaningful information from collected big data and visualized it. Inferring users’ reviews, figuring out opinions objectively and seeking ways to improve games are helpful in improving mobile RPG that can be played continuously.

(1) Background: The CEO message of hospital homepage contain various contents such as the hospital's future vision, promises with customers, upgraded services and public activities. The CEO’s message of the homepage includes non-financial information as well as financial information of corporates. Also, it provides useful information for not only company's goals and vision but also firm performance and strategies for the future. This study aims to investigate associations between CEO’s message of hospitals homepages and financial status. We used the balanced scorecard frame to analyze what content on the hospital's homepage is related to the hospital's various financial ratios. (2) Methods: We adopt a text mining method to extract significantly repeated keywords from the CEO’s message of hospital website. And we classify these keywords by a balanced scorecard frame. To examine the relationship between keywords of CEO’s message of the hospital homepage and hospital’s financial ratio, T-test is conducted for the difference in the TF-IDF (Term Frequency is Divided by Inverse Document Frequency) mean of the home page contents and its relationship with the views of the balanced scorecard framework. (3) Results: According to empirical results on 65 samples collected from local hospitals, there are some significant relationship between the qualitative content of the hospital's homepage and the quantitative financial ratio that indicates profitability, activity, leverage, liquidity, and transfer to essential business fund (EBF) income. (4) Conclusions: The introduction section of a homepage is most accessible to customers, containing the aims and ideals of hospitals and reflecting their values and visions [1]. In addition, in view of financial status, they can either emphasize financial strength or focus on other areas to mask weakness of financial information. This study reminds us of the importance of hospital website’s disclosure, and it can be inferred from the financial status of the hospital. It also highlights the need for harmonization between quantitative data, financial statements, and qualitative data, CEO’s messages. (5) Implications: To our best knowledge, this paper is the first research attempting to investigate the relation between text of hospital homepage and financial ratio of hospital through text mining technique and balanced scorecard frame. Hospitals take a crucial part in a country’s welfare and healthcare backbone industry. Nevertheless, in many countries, hospital organization sectors tend to remain a source of critical fiscal deficits due to its ineffective and sloppy management. We expect that the result of this paper can provide hospital managers to useful information.

Manual review of the extensive literature covering nutrition-based lifestyle interventions to promote healthy cognitive ageing has proved educative, however, data-driven techniques can better account for the large size of the literature (tens of thousands of potentially relevant publications to date) and interdisciplinary nature of where relevant publications may be found. In this study we present a new way to map the literature landscape focusing on nutrition-based lifestyle interventions to promote healthy cognitive ageing. We applied a combination of citation network analysis and text mining to map out the existing literature on nutritional interventions and cognitive health. Results indicated five overarching clusters of publications, which could be further deconstructed into a total of 35 clusters. These could be broadly distinguished by focus on lifespan stages (e.g. infancy versus older age), and specificity regarding nutrition (e.g. narrow focus on iodine deficiency versus broad focus on weight gain). Rather than concentrating into a single cluster, interventions were present throughout the majority of the research. We conclude that a data-driven map of the nutritional intervention literature can benefit the design of future interventions, by highlighting topics and themes that could be synthesized across currently disconnected clusters of publications.

In an era when medical literature is increasing daily, researchers in biomedical and clinical areas have joined efforts with language engineers to analyze large amount of biomedical and molecular biology literature (such as PubMed), patient data or health records. With such a huge amount of reports, evaluating their impact has long seized to be a trivial task. In this context, this paper intends to introduce a non-scientific factor that represents an important element in the effort of gaining acceptance of claims. Thus, we postulate that the confidence the author is expressing in his work plays an important role in shaping the first impression that influences the reader’s perception of the paper. The results discussed in this paper are based on a series of experiments ran over data from the Open Archives Initiative (OAI) corpus that provides interoperability standards in order to facilitate the effectiveness dissemination of the content. This method can be useful to the direct beneficiaries (authors, who are engaged in medical or academic research), but, also, researchers in the fields of BioNLP and NLP, etc.

The research presents the Methodology of Improving the Accuracy in Text Classification in Light of Modelling the Latent Semantic Relations (LSR). The aim of this Methodology is to find the ways of eliminating the Limitations of Discriminant and Probabilistic methods for LSR revealing and customizing the Text Classification Process to the more accurate recognition of the text tonality. This aim should be achieved by using the knowledge about the text’s Hierarchical Semantic Context in the form of Corpora-based Hierarchical Sentiment Dictionary. The main scientific contribution of this research is the following set of approaches to improve the qualitative characteristics of Text Classification process: combination of the Discriminant and Probabilistic methods allowing to decrease the influences of the Limitations of these methods on the LSR revealing process; considering each document as a complex structure allowing to estimate documents integrally by separated classification of topically completed textual component (paragraphs); taking into account the features of Argumentative type of documents (Reviews) allowing to use the author’s subjective evaluation of text tonality for development the Text Classification methodology. Tonality, expressed by the Review’s author, has a significant, but not critical, effect on the qualitative indicators of Sentiment Recognition.

This study mostly analyzes the pragmatic viewpoints of speech acts and language functions through Halliday’s (1975) language functions and Searle’s (1976) speech acts were adapted to analyze the functional aspects of the conversations and it was also intended to explore the teachers’ perception toward teaching and learning English as a Foreign Language (EFL) learners’ textbooks and the teacher’s components of communicative knowledge regarding the functions of language in daily activity. The participants of this paper consisted of thirteen Sunrise 10, 11, 12 grades of Kurdish teachers at high school English in Iraqi Kurdistan. Through semi-structured interview, it was found that the conversations in the mentioned textbooks are insufficient from the pragmatic point of view. Some recommendation for the textbook designers, teachers, material developer were raised to make up the shortcomings of the textbooks. The findings reveal that the conversation texts in Sunrise textbooks are not meeting the systematic standard of pragmatic competence the English language learners and the book designer must be recommended to be aware of those shortcomings of the textbook series if they are required to develop their speaking skills in both student and activity series. The implications of this paper can be helpful in comparing the results of this study with other similar studies to check if there is a universal pattern in performing the speech acts and language functions and the interest to Kimberley Education for Life learners in increasing their knowledge of pragmatics in general and the role of language functions and speech acts investigated in this study.

Purpose: This study aimed to examine how aging and modifications of critical acoustic parameters may affect the perception of whispered speech as a degraded signal. Method: Forty Mandarin-speaking adults were included in the study. Part 1 of the study compared the perception of Mandarin lexical tones, vowels, and syllables in older and younger adults in whispered vs. phonated speech conditions. Parts 2 and 3 further examined how modification of duration and intensity cues contributed to the perceptual outcomes. Results: Perception of whispered tones was compromised in older and younger adults. Older adults identified lexical tones less accurately than their younger counterparts, particularly for phonated T2, T3 and whispered T3. Aging also negatively affected the vowel identification of /i, u/ in the whispered condition. Syllable-level accuracy was largely dependent on the accuracy of lexical tones and vowels. Furthermore, reduced duration led to the decreased accuracy of phonated T3 and whispered T2, T3 but increased accuracy of phonated T4. Reduced intensity lowered the recognition accuracy for phonated vowels /i, ɤ, o, y/ in older adults and /i, u/ in younger adults, and it also lowered the accuracy of whispered vowels /a, ɤ/ in older adults. Contrary to our expectation, increased duration and intensity did not improve older adults’ speech perception in either phonated or whispered conditions. Conclusion: The results suggest that aging adversely affected speech perception in both phonated and whispered conditions with more challenges in identifying whispered speech for older adults. While older adults’ diminished performance may be potentially due to problems with processing the degraded temporal and spectral information of the target speech sounds, it cannot be simply compensated for by increasing the duration and intensity of the target sounds beyond the audible level.

How language mediates emotional perception and experience is poorly understood. The present event-related potential (ERP) study examined the explicit and implicit processing of emotional speech to differentiate the relative influences of communication channel, emotion category and task type in the prosodic salience effect. Thirty participants (15 women) were presented with spoken words denoting happiness, sadness and neutrality in either the prosodic or semantic channel. They were asked to judge the emotional content (explicit task) and speakers’ gender (implicit task) of the stimuli. Results indicated that emotional prosody (relative to semantics) triggered larger N100 and P200 amplitudes with greater delta, theta and alpha inter-trial phase coherence (ITPC) values in the corresponding early time windows, and continued to produce larger LPC amplitudes and faster responses during late stages of higher-order cognitive processing. The relative salience of prosodic and semantics was modulated by emotion and task, though such modulatory effects varied across different processing stages. The prosodic salience effect was reduced for sadness processing and in the implicit task during early auditory processing and decision-making but reduced for happiness processing in the explicit task during conscious emotion processing. Additionally, across-trial synchronization of delta, theta and alpha bands predicted the ERP components with higher ITPC values significantly associated with stronger N100, P200 and LPC enhancement. These findings reveal the neurocognitive dynamics of emotional speech processing with prosodic salience tied to stage-dependent emotion- and task-specific effects, which can reveal insights to research reconciling language and emotion processing from cross-linguistic/cultural and clinical perspectives.

When participants in a statistical learning paradigm are asked to learn from two incompatible or competing inputs, they often fail to learn from one or both inputs. This study presents the results of two experiments that were both completed by one group of typically developing four-year-old children. One experiment targeted word-medial consonant patterns (phonotactics), whereas the other targeted strong-weak and weak-strong stress patterns (prosody). The order of the experiments was critical for learning outcomes in the phonotactics experiment: When children learned phonotactics first, their production accuracy increased following exposure to a high frequency input. When children learned phonotactics second, however, their production accuracy dropped when they were exposed to the high frequency input. Results from the prosody experiment were inconclusive, with limited evidence of any learning effect. Overall, the results suggest that children may conflate learning experiences, and patterns learned from an initial experimental input compete with patterns in a subsequent experiment. When considering natural language acquisition, the results suggest that an isolated episode of learning may lead to generalizations that are incompatible with later input, and possibly, with larger patterns in the language.

This paper explores the suppression of ideas within academic scholarship by academics, either by self-suppression or because of the efforts of other academics. Legal, moral, and social issues distinguishing freedom of speech, freedom of inquiry, and academic freedom are reviewed. How these freedoms and protections can come into tension is then explored by an analysis of denunciation mobs who exercise their legal free speech rights to call for punishing scholars who express ideas they disapprove of and condemn. When successful, these efforts, which constitute legally protected speech, will suppress certain ideas. Real-world examples over the past five years of academics who have been sanctioned or terminated for scholarship targeted by a denunciation mob are then explored.

Long unpunctuated texts containing complex linguistic sentences are a stumbling block to processing any low-resource languages. Thus, approaches that attempt to segment lengthy texts with no proper punctuation into simple candidate sentences are a vitally important preprocessing task in many hard-to-solve NLP applications. In this paper, we propose (PDTS) a punctuation detection approach for segmenting Arabic text, built on top of a multilingual BERT-based model and some generic linguistic rules. Furthermore, we showcase how PDTS can be effectively employed as a text tokenizer for unpunctuated documents (i.e., mimicking the transcribed audio-to-text documents). Experimental findings across two evaluation protocols (involving an ablation study and a human-based judgment) demonstrate that PDTS is practically effective in both performance quality and computational cost.

With the development of deep learning, several Graph Neural Networks (GNN)-based approaches have been utilized for text classification. However, GNNs encounter challenges in capturing contextual text information within a document sequence. To address this, a novel text classification model RB-GAT is proposed by combining RoBERTa-BiGRU embedding and a multi-head Graph ATtention Network (GAT). First, the pre-trained RoBERTa model is exploited to learn word and text embeddings in different contexts. Second, the Bidirectional Gated Recurrent Unit (BiGRU) is employed to capture long-term dependencies and bidirectional sentence information from the text context. Next, the multi-head graph attention network is applied to analyze this information, which serves as a node feature for the document. Finally, the classification results are generated through a Softmax layer. Experimental results on three benchmark datasets demonstrate that our method can achieve an accuracy of 71.48%, 98.45%, and 80.32% on Ohsumed, R8, and MR, which is superior to the existing nine text classification approaches.

In today's world, news events are emerging incessantly. Classifying the same event reported by different countries is of significant importance for public opinion control and intelligence gathering. Due to the diverse types of news, relying solely on translators would be costly and inefficient, while depending solely on translation systems would incur considerable performance overheads in invoking translation interfaces and storing translated texts. To address this issue, we mainly focus on the clustering problem of cross-lingual news. To be specific, we use a combination of sentence vector representations of news headlines in a mixed semantic space and the topic probability distributions of news content to represent a news article. In the training of cross-lingual models, we employ knowledge distillation techniques to fit two semantic spaces into a mixed semantic space. We abandon traditional static clustering methods like K-Means and AGNES in favor of the incremental clustering algorithm Single-Pass, which we further modify to better suit cross-lingual news clustering scenarios. Our main contributions are as follows: (1) We adopt the English standard BERT as the teacher model and XLM-Roberta as the student model, training a cross-lingual model through knowledge distillation that can represent sentence-level bilingual texts in both Chinese and English. (2) We use the LDA topic model to represent news as a combination of cross-lingual vectors for headlines and topic probability distributions for content, introducing concepts such as topic similarity to address the cross-lingual issue in news content representation. (3) We adapt the Single-Pass clustering algorithm for the news context to make it more applicable. Our optimizations of Single-Pass include adjusting the distance algorithm between samples and clusters, adding cluster merging operations, and incorporating a news time parameter.

Text summarization remains a challenging task in the Natural Language Processing field despite the plethora of applications in enterprises and daily life. One of the common use cases is the summarization of web pages which has the potential to provide an overview of web pages to devices with limited features. In fact, despite the increasing penetration rate of mobile devices in rural areas, the bulk of those devices offer limited features in addition to the fact that these areas are covered with limited connectivity such as the GSM network. Summarizing web pages into SMS becomes, therefore, an important task to provide information to limited devices. This work introduces WATS-SMS, a T5-based French Wikipedia Abstractive Text Summarizer for SMS. It is built through a transfer learning approach. The T5 English pre-trained model is used to generate a French text summarization model by retraining the model on 25,000 Wikipedia pages then compared with different approaches in the literature. The objective is twofold: (1) to check the assumption made in the literature that abstractive models provide better results compared to extractive ones; and (2) to evaluate the performance of our model compared to other existing abstractive models. A score based on ROUGE metrics gave us a value of 52% for articles with length up to 500 characters against 34.2% for transformer-ED and 12.7% for seq-2seq-attention; and a value of 77% for articles with larger size against 37% for transformers-DMCA. Moreover, an architecture including a software SMS-gateway has been developed to allow owners of mobile devices with limited features to send requests and to receive summaries through the GSM network.

In the present, the growth of digitization and worldwide communications make OCR systems of exotic languages a very important task. In this paper, we attempt to develop an OCR system for one of these exotic languages with a unique script, Amharic. Motivated by the recent success of the Attention mechanism in Neural Machine Translation (NMT), we extend the attention mechanism for Amharic text-image recognition. The proposed model consists of CNNs and attention embedded recurrent encoder-decoder networks that are integrated following the configuration of the seq2seq framework. The attention network parameters are trained in an end-to-end fashion and the context vector is injected, with the previously predicted output, at each time steps of decoding. Unlike the existing OCR model that minimizes the CTC objective function, the new model minimizes the categorical cross-entropy loss. The performance of the proposed attention-based model is evaluated against the test dataset from the ADOCR database which consists of both printed and synthetically generated Amharic text-line images and achieved promising results with a CER of 1.54% and 1.17% respectively.

This paper propose an original approach of achieving a Cymatics based visual perception of image-extracted text. In this context, an effective approach for automated text detection and recognition for the natural scene images is proposed. The incoming image is firstly enhanced by employing CLAHE and DWT. Afterwards, the text regions of the enhanced image are detected by employing the MSER feature detector. The non-text MSERs are removed by employing the geometrical and contour based filters. The remaining MSERs are grouped into words or phrases by finding out similarities between them. The text recognition is performed by employing an OCR function. The extracted text is sequentially analysed on character by character basis. Each character is converted into a methodical acoustic excitation. Finally, these excitations are converted into the systematic visual perceptions by using the phenomenon of Cymatics. The system functionality is tested with an experimental setup. For the case of studied natural scenes, the suggested approach achieves 80% precision in text localization and 53% precision in end-to-end text recognition. The devised system principle is novel and can be employed in various applications like visual art, encryption, education, integration of impaired people, etc.

In recent years, with the popularity of smart mobile devices, the interaction between devices and users, especially in the form of voice interaction, has become increasingly important. If smart devices can understand more users' emotional states through voice data, more customized services can be provided for users. This paper proposes a novel machine learning model for speech emotion recognition, which combines convolutional neural networks (CNN), long short-term memory neural networks (LSTM), and deep neural networks (DNN), called CLDNN. To make the designed system can recognize the audio signal closer to the human auditory system does, this article uses the Mel frequency cepstral coefficients (MFCCs) of audio data as the input of the machine learning model. First, the MFCCs of the voice signal is extracted as the input of the model, and the feature values of the data are calculated using several local feature learning blocks (LFLB) composed of one-dimensional CNN. Because the audio signals are time-series data, the feature values obtained from LFLBs then input into LSTM layer to enhance the learning on time-series level. Finally, fully connected layers are used for classification and prediction. Three databases RAVDESS, EMO-DB and IEMOCAP are used for the experiments in this paper. The experimental results show that the proposed method can improve the accuracy compared to other related researches in speech emotion recognition.

Potocki-Lupski Syndrome (PTLS) is a rare condition associated with a duplication of 17p11.2 that may underlie a wide range of congenital abnormalities and heterogeneous behavioral phenotypes. Along with developmental delay and intellectual disability, autism-specific traits are often reported to be the most common among patients with PTLS. To contribute to the discussion of the role of autism spectrum disorder (ASD) in the PTLS phenotype, we present a case of a female adolescent with a de novo dup(17)(p11.2p11.2) without ASD features, focusing on in-depth clinical, behavioral, and electrophysiological (EEG) evaluations. Among EEG features, we found the atypical peak-slow wave patterns and a unique saw-like sharp wave of 13 Hz that was not previously described in any other patient. The power spectral density of the resting state EEG was typical in our patient with only the values of non-linear EEG dynamics: Hjorth complexity and Fractal dimension were drastically attenuated compared with the patient’s neurotypical peers. Here we also summarize results from previously published reports of PTLS that point to the about 21% occurrence of ASD in PTLS that might be biased, taking into account methodological limitations. More consistent among PTLS patients were intellectual disability and speech and language disorders.

The growth in online child exploitation material is a significant challenge for European Law Enforcement Agencies (LEAs). One of the most important sources of such online information corresponds to audio material that needs to be analyzed to find evidence in a timely and practical manner. That is why LEAs require a next-generation AI-powered platform to process audio data from online sources. We propose the use of speech recognition and keyword spotting to transcribe audiovisual data and to detect the presence of keywords related to child abuse. The considered models are based on two of the most accurate neural-based architectures to date: Wav2vec2.0 and Whisper. The systems are tested under an extensive set of scenarios in different languages. Additionally, keeping in mind that obtaining data from LEAs is very sensitive, we explore the use of federated learning to have more robust systems for the addressed application, while maintaining the privacy of the data to LEAs. The considered models achieved a word error rate between 11% and 25%, depending on the language. In addition, the systems are able to recognize a set of spotted words with true positives rates between 82% and 98%, depending on the language. Finally, federated learning strategies show that they can maintain and even improve the performance of the systems when compared to centralized trained models. The proposed systems sit the basis for an AI-powered platform for automatic analysis of audio in the context of forensic applications within child abuse. The use of federated learning is also promising for the addressed scenario, where data privacy is an important issue to be managed.

The promised dataset was obtained from the daily Turkish words and phrases pronounced by various people in the videos posted on YouTube. The purpose of collecting the dataset is to provide detection of the spoken word by recognizing patterns or classifying lip movements with supervised, unsupervised, semi-supervised learning and machine learning algorithms. Most of the datasets related with lip reading consist of people recorded on camera with fixed backgrounds and the same conditions, but the dataset presented here consists of images compatible with machine learning models developed for real-life challenges. It contains a total of 2335 instances taken from TV series, movies, vlogs, and song clips on YouTube. The images in the dataset vary due to factors such as the way people say words, accent, speaking rate, gender and age. Furthermore, the instances in the dataset consist of videos with different angles, shadows, resolution, and brightness that are not created manually. The most important feature of our lip reading dataset is that we contribute to the non-synthetic Turkish dataset pool, which does not have wide dataset varieties. Machine learning studies can be carried out in many areas, such as the defense industry and social life, with this dataset.

Data augmentation techniques recently gained more adoption in speech processing, including speech emotion recognition. Although more data tends to be more effective, there may be a trade-off in which more data will not provide a better model. This paper reports experiments on investigating the effects of data augmentation in speech emotion recognition. The investigation aims at finding the most useful type of data augmentation and the number of data augmentations for speech emotion recognition. The experiments are conducted on the Japanese Twitter-based emotional speech corpus. The results show that for speaker-independent data, two data augmentations with glottal source extraction and silence removal exhibited the best performance among others, even with more data augmentation techniques. For the text-independent data (including speaker and text-independent), more data augmentations tend to improve speech emotion recognition performances. The results highlight the trade-off between the number of data augmentation and the performance of speech emotion recognition showing the necessity to choose a proper data augmentation technique for a specific application.

We present improvements in n-best rescoring of code-switched speech achieved by n-gram augmentation as well as optimised pretraining of long short-term memory (LSTM) language models with larger corpora of out-of-domain monolingual text. In addition, we consider the application of large pretrained transformer-based architectures. Our experimental evaluation is performed on an under-resourced corpus of code-switched speech comprising four bilingual code-switched sub-corpora, each containing a Bantu language (isiZulu, isiXhosa, Sesotho, or Setswana) and English. We find in our experiments that, by combining n-gram augmentation with the optimised pretraining strategy, speech recognition errors are reduced for each individual bilingual pair by 3.51% absolute on average over the four corpora. Importantly, we find that even speech recognition at language boundaries improves by 1.14% even though the additional data is monolingual. Utilising the augmented n-grams for lattice generation, we then contrast these improvements with those achieved after fine-tuning pretrained transformer-based models such as distilled GPT-2 and M-BERT. We find that, even though these language models have not been trained on any of our target languages, they can improve speech recognition performance even in zero-shot settings. After fine-tuning on in-domain data, these large architectures offer further improvements, achieving a 4.45% absolute decrease in overall speech recognition errors and a 3.52% improvement over language boundaries. Finally, a combination of the optimised LSTM and fine-tuned BERT models achieves a further gain of 0.47% absolute on average for three of the four language pairs compared to M-BERT. We conclude that the careful optimisation of the pretraining strategy used for neural network language models can offer worthwhile improvements in speech recognition accuracy even at language switches, and that much larger state-of-the-art architectures such as GPT-2 and M-BERT promise even further gains.

Users on the social networking platform have the freedom to express themselves freely. Towards the same time, this has created a forum for disagreement and hate directed at someone, society, racism, sexual orientation, and so on. Identifying hate online is a challenging task. Researchers from all around the world have contributed major methods for detecting hate speech, but owing to the issue's complexity, there are still many unresolved issues. In this research, we offer a multi-model learning strategy for detecting hate speech on Twitter. We utilised the Kaggle TwitterHate dataset, which had 31962 tweets categorised as binary hate or non-hate, to evaluate our technique. The suggested method is tested using commonly used machine learning classifiers with multi-model technique. Using TF-IDF features, we acquired detection results of 96.29 %, precision of 96%, recall of 96%, and f1-score of 96%.

The main restriction of the Semantic Web is the difficult of the SPARQL language, that is necessary to extract information from the Knowledge Representation also known as ontology. Making the Semantic Web accessible for people who do not know SPARQL, is essential the use of friendlier interfaces and a good alternative is Natural Language. This paper shows the implementation of a friendly prototype interface to query and retrieve, by voice, information from website building with the Semantic Web tools. In that way, the end users avoid the complicated SPARQL language. To achieve this, the interface recognizes a speech query and converts it into text, it processes the text through a java program and identifies keywords, generates a SPARQL query, extracts the information from the website and read it in voice, for the user. In our work Google Cloud Speech API makes Speech-to-Text conversions and Text-to Speech conversions are made with SVOX Pico. As results, we have measured three variables: The success rate in queries, the response time of query and a usability survey. The values of the variables allows the evaluation of our prototype. Finally the interface proposed provides us a new approach in the problem, using the Cloud like a Service, reducing barriers of access to the Semantic Web for people without technical knowledge of Semantic Web technologies.

Purpose of publication: - Preparation of bibliometric data exported from the Scilit platform on energy efficiency and conservation for further analysis to identify relevant research topics. - To identify potential issues in the processing of data exported from the Scilit platform. - Providing colleagues with the opportunity to use the prepared data and examples of their analysis for independent research on topical issues of energy efficiency and energy conservation using materials provided by the Scilit platform. Research Materials: Files in CSV and RIS formats exported from Scilit for the query "energy conservation OR efficiency" in Common Fields [Title, Abstract, Keyword], using filters: Content Type → Journal Article; Year→2021-2023; Subject → Industrial Engineering (29.8K), Energy and Fuel Technology (9.8K), Manufacturing Engineering (9.2K). A total of 30K records sorted by their relevance (10K for each year) were exported. Data are current as of 14-03-2024. Methods: Preprocessing of title, annotation, and keyword field texts using lemmatization dictionaries collected on GitHub, removal of keywords taken from GATE and spaCy, and "manual" editing. Using VOSviewer to analyze publication topics by clustering keywords based on their co-occurrence. Using Scimago Graphica to build bubble diagrams.Application of the GSDMM algorithm for clustering bibliometric records by title and annotation texts. Creation of a dictionary for this algorithm using the keyword field.Use of the Carrot2 demo version and the NMF algorithm for a more detailed analysis of the topics of the record clusters obtained from GSDMM. Results: are presented in the form of initial and interim tables and graphs obtained in the course of this study. The full tables are provided as references to the attached materials. Supplementary material for this preprint on figshare: Chigarev, Boris (2024). Supplementary material for preprint "Analyzing the Possibilities of Using the Scilit Platform to Identify Current Energy Efficiency and Conservation Issues". figshare. Dataset. https://doi.org/10.6084/m9.figshare.25574058.v1

Text mining is a technological trend often highlighted in the continuous exchange of information through interconnected media. Its applicability goes beyond private organizations, as the public sector also requires it to treat textual information regarding services offered. Within this scenario, public security emerges as a prominent user of text mining that seeks to ensure the construction of data and knowledgebases to support decision-making about law enforcement actions to ensure citizen welfare. The primary objectives of this article are: (i) to develop a survey to identify text mining applications, techniques, opportunities, and challenges in public security, and (ii) to outline research directions concerning these topics and provide insights so that interested researchers can develop new studies. The literature was searched within four databases: Scopus, IEEE Xplore, ACM Digital Library, and Web of Science. A filtering process was applied to extract the works most aligned with the target theme, resulting in the selection of 194 of the most relevant works for a literature review. There were identified nineteen key applications of text mining related to public security and the most recurrent techniques and technologies reported between 2014 to 2021, supporting outlining three axes for future directions: one with possible expansion of objectives for new research; another on changes and adaptations in scopes for the methodological context; and the last one on expansions and changes in application scenarios based on the literature.

(1) Background: Breast cancer is a frequent heterogeneous disorder diagnosed in woman and is a high cause of mortality of them in reason to rapid metastasis and disease recurrence. Ferroptosis can inhibit breast cancer cell growth, improve the sensitivity of chemotherapy and radiotherapy and inhibit distant metastases so potentially acts on tumor micro-environment; (2) Methods: Ferroptosis/Extracellular matrix remodeling literature text-mining results were integrated in breast cancer transcriptome cohort according their distant relapse free survival (DRFS) under adjuvant therapy (anthracyclin+taxanes) and also in MDA-MB-231 transcriptome functional experiments with ferroptosis activations (GSE173905); (3) Results: Ferroptosis/Extracellular matrix remodeling text-mining identified 910 associated genes in at list 10 articles. Univariate Cox analyses censored on breast cancer (GSE25066) selected 252 individual significant genes and 171 of them found with an adverse expression. Functional enrichment of these 171 adverse genes predicted basal breast cancer signatures. By text-mining some ferroptosis significant adverse selected genes shared citations in domain of ECM remodeling such as: TNF, IL6, SET, CDKN2A, EGFR, HMGB1, KRAS, MET, LCN2, HIF1A, TLR4. A molecular score based on expression the eleven genes was found predictive of worst prognosis breast cancer at univariate level: basal subtype, short DRFS, high grade values 3 and 4, estrogen and progesterone receptors negative and nodal stages 2 and 3. This eleven gene signature was validated as regulated by ferroptosis inductors (erastin and RSL3) in triple negative breast cancer cellular model MDA-MB-231.; (4) Conclusions: Crosstalk between ECM remodeling-Ferroptosis functionalities allowed to define a molecular score which have been characterized as an independent adverse parameter in prognosis of breast cancer patients. Gene signature of this molecular score have been validated to be regulated by erastin/RSL3 ferroptosis activators. This molecular score could be promising to evaluate ECM impact of ferroptosis target therapies in breast cancer.

The body of scientific literature continues to grow annually. Over 1.5 million abstracts of biomedical publications were added to the PubMed database in 2021. Therefore, developing cognitive systems that provide a specialized search for information in scientific publications based on subject area ontology and modern artificial intelligence methods is urgently needed. We previously developed a web-based information retrieval system, ANDDigest, designed to search and analyze information in the PubMed database using a customized domain ontology. This paper presents an improved ANDDigest version that uses fine-tuned PubMedBERT classifiers to enhance the quality of short name recognition for molecular-genetics entities in PubMed abstracts on eight biological object types: cell components, diseases, side effects, genes, proteins, pathways, drugs, and metabolites. This approach increased average short name recognition accuracy by 13%. The new ANDDigest version (01.2022) has a web interface and is freely available to users at https://anddigest.sysbio.ru/.

The spread of the Coronavirus pandemic has been accompanied by an infodemic. The false information that is embedded in the infodemic affects people’s ability to have access to safety information and follow proper procedures to mitigate the risks. This research aims to target the falsehood part of the infodemic, which prominently proliferates in news articles and false medical publications. Here, we present NeoNet, a novel supervised machine learning text mining algorithm that analyzes the content of a document (news article, a medical publication) and assigns a label to it. The algorithm is trained by TFIDF bigram features which contribute a network training model. The algorithm is tested on two different real-world datasets from the CBC news network and Covid-19 publications. In five different fold comparisons, the algorithm predicted a label of an article with a precision of 97-99 %. When compared with prominent algorithms such as Neural Networks, SVM, and Random Forests NeoNet surpassed them. The analysis highlighted the promise of NeoNet in detecting disputed online contents which may contribute negatively to the COVID-19 pandemic.

A global event such as the COVID-19 crisis presents new, often unexpected responses that are fascinating to investigate from both, scientific and social standpoints. Despite several documented similarities, the Coronavirus pandemic is clearly distinct from the 1918 flu pandemic in terms of our exponentially increased, almost instantaneous ability to access/share information, offering an unprecedented opportunity to visualise rippling effects of global events across space and time. Personal devices provide “big data” on people’s movement, the environment and economic trends, while access to the unprecedented flurry in scientific publications and media posts provides a measure of the response of the educated world to the crisis. Most bibliometric (co-authorship, co-citation, or bibliographic coupling) analyses ignore the time dimension, but COVID-19 has made it possible to perform a detailed temporal investigation into the pandemic. Here, we report a comprehensive network analysis based on more than 20000 published documents on viral epidemics, authored by over 75,000 individuals from 140 nations in the past one year of the crisis. In contrast to the 1918 flu pandemic, access to published data over the past two decades enabled a comparison of publishing trends between the ongoing COVID-19 pandemic and those of the 2003 SARS epidemic, to study changes in thematic foci and societal pressures dictating research over the course of a crisis.

This study aims to understand the global environment of COVID-19 management and guide future policy directions after the pandemic crisis. To this end, we analyzed a series of the World Economic Forum’s COVID-19 response reports through text mining and network analysis. These reports, written by experts in diverse fields, discuss multidimensional changes in socioeconomic situations, various problems created by those changes, and strategies to respond to national crises. Based on 3,897 refined words drawn from a morphological analysis of 26 reports (as of the end of 2020), this study analyzes the frequency of words, the relationships among words, the importance of specific documents, and the connection centrality through text mining. In addition, network analysis helps develop strategies for sustainable response to and management of national crises through identifying clusters of words with a similar structural equivalence.

Topological Data Analysis (TDA) refers to a collection of methods that find the structure of shapes in data. Although recently, TDA methods have been used in many areas of data mining, it has not been widely applied to text mining tasks. In most text processing algorithms, the order in which different entities appear or co-appear is being lost. Assuming these lost orders are informative features of the data, TDA may play a significant role in the resulted gap on text processing state of the art. Once provided, the topology of different entities through a textural document may reveal some additive information regarding the document that is not reflected in any other features from traditional text processing methods. In this paper, we introduce a novel approach that hires TDA in text processing in order to capture and use the topology of different same-type entities in textural documents. First, we will show how to extract some topological signatures in the text using persistent homology-i.e., a TDA tool that captures topological signature of data cloud. Then we will show how to utilize these signatures for text classification.

The aim of this study is to describe the cognitive and speech results obtained after growth hormone (GH) treatment and neurorehabilitation in a man that suffered a traumatic brain injury (TBI). 17 months after the accident, the patient was treated with growth hormone (GH), together with neurostimulation and speech therapy. At admission, the left vocal cord revealed paralyzed, in the paramedian position, a situation compatible with a recurrent nerve injury. Clinical and rehabilitation assessments revealed a prompt improvement in speech and cognitive functions, and following completion of treatment, endoscopic examination showed recovery of vocal cord mobility. These results, together with previous results from our group, indicate that GH treatment is safe and effective for helping neurorehabilitation in chronic speech impairment due to central laryngeal paralysis, as well as impaired cognitive functions.

Several researchers have contemplated deep learning-based post-filters to increase the quality of statistical parametric speech synthesis, which perform a mapping of the synthetic speech to the natural speech, considering the different parameters separately and trying to reduce the gap between them. The Long Short-term Memory (LSTM) Neural Networks have been applied successfully in this purpose, but there are still many aspects to improve in the results and in the process itself. In this paper, we introduce a new pre-training approach for the LSTM, with the objective of enhancing the quality of the synthesized speech, particularly in the spectrum, in a more efficient manner. Our approach begins with an auto-associative training of one LSTM network, which is used as an initialization for the post-filters. We show the advantages of this initialization for the enhancing of the Mel-Frequency Cepstral parameters of synthetic speech. Results show that the initialization succeeds in achieving better results in enhancing the statistical parametric speech spectrum in most cases when compared to the common random initialization approach of the networks.

The IoT Internet of Things being a promising technology of the future. It is expected to connect billions of devices. The increased communication number is expected to generate data mountain and the data security can be a threat. The devices in the architecture are fundamentally smaller in size and low powered. In general, classical encryption algorithms are computationally expensive and this due to their complexity and needs numerous rounds for encrypting, basically wasting the constrained energy of the gadgets. Less complex algorithm, though, may compromise the desired integrity. In this paper we apply a lightweight encryption algorithm named as Secure IoT (SIT) to a quantized speech image for Secure IoT. It is a 64-bit block cipher and requires 64-bit key to encrypt the data. This quantized speech image is constructed by first quantizing a speech signal and then splitting the quantized signal into frames. Then each of these frames is transposed for obtaining the different columns of this quantized speech image. Simulations result shows the algorithm provides substantial security in just five encryption rounds.

This paper presents a novel approach to sentiment analysis specifically customized for predicting stock market movements, bypassing the need for external dictionaries which are often unavailable for many languages. Our methodology directly analyzes textual data, with a particular focus on context-specific sentiment words within neural network models. This specificity ensures our sentiment analysis is both relevant and accurate in identifying trends in the stock market. We employ sophisticated mathematical modeling techniques to enhance both the precision and interpretability of our models. Through meticulous data handling and advanced machine learning methods, we leverage large datasets from Twitter and financial markets to examine the impact of social media sentiment on financial trends. We achieved an accuracy exceeding 75%, highlighting the effectiveness of our modeling approach, which we further refined into a convolutional neural network model. This achievement contributes valuable insights into sentiment analysis within the financial domain, thereby improving the overall clarity of forecasting in this field.

This study focuses on understanding how global auto parts companies manage innovation by analyzing keywords in CEO messages from their reports. The study involved three key phases: (1) extraction of innovation-related keywords emphasized in CEO messages, (2) comparison of innovation-related keyword usage between high-performing (upper group) and low-performing (lower group) companies, and (3) analysis of the characteristics of innovation management via internal meaning and the structural interpretation of keywords. A comparison between the upper and lower groups revealed distinct differences in the emphasized keywords. The mapping of open innovation success factors, based on keywords extracted according to eigenvector centrality, identified Provision of Resources and Governance as central to innovation management in the upper group. These findings serve as a benchmark for establishing the direction, vision, and strategy of innovation management in the lower group.