Teaching the Sustain Pedal in the Digital Studio: Linking Pedal Action, Acoustic Result, and Pedagogical Feedback

2. Introduction

Students usually meet the sustain pedal through one simple instruction: press the right pedal so the sound keeps going after the fingers lift off the keys. That description works fine at an elementary level. Musically and pedagogically, though, it papers over several distinct events. A key can be released while its string keeps sounding. A pedal can be partially depressed rather than fully down. Dampers can interact with strings mid-release. Sympathetic resonance can reshape the spectrum of the whole instrument. And whether the resulting sonority actually works depends on harmony, texture, tempo, articulation, instrument, room, repertoire, and interpretive intention — all at once. Treat every one of these as a single variable called "sustain," and action, sound, and musical judgment collapse into one undifferentiated category.

Performance research and pedagogy have recognized the complexity of pedal technique for a long time. Heinlein (1929) used experimental methods to study individual differences in damper-pedal performance and treated pedaling as a vehicle for interpretive individuality — not a fixed mechanical routine. Repp (1996, 1997) later showed that pedal timing relative to key events doesn't follow one invariant rule across tempi or performers. Lipke-Perry et al. (2022), more recently, found that pianists' pedaling shifts under different attentional-focus conditions, with the patterns tied to expertise and preparation. In the pedagogical literature, Al Bakri (2025) documents ongoing terminological disagreement across piano schools and catalogs thirteen distinct forms or techniques associated with the right pedal. All of this points to the same basic fact: skilled pedaling doesn't reduce to a fixed on/off recipe.

The acoustic mechanism is just as complicated. Once the sustain pedal engages, dampers lift off the strings, decay behavior changes, and sympathetic vibration becomes more likely (Lehtonen et al., 2007). Part-pedaling adds further states, in which damper-string interaction shifts both decay and timbre depending on pedal depth (Lehtonen et al., 2009). So the physical state of a pedal and the acoustic state of the instrument are related — but they are not the same description of the same phenomenon, and treating them as interchangeable is where trouble starts.

Digital pianos, MIDI recording, sensor systems, and audio-analysis tools now make pedal behavior visible in ways that used to be hard to inspect at all. Earlier computational work used MuSA.RT to model and visualize how sustain-pedal use shifted tonal coherence in a controlled piano performance, explicitly flagging relevance to piano pedagogy and automated transcription (Chew & François, 2008). Liang et al. (2017) built the Piano Pedaller system to capture physical pedal gestures, classify techniques, and display results inside a score-following environment. Liang et al. (2018) extended that work to measurement, recognition, and visualization of pedaling gestures and techniques; later work moved on to automatic sustain-pedal detection in polyphonic audio (Liang et al., 2019). Technology-mediated piano teaching has also drawn on visual and MIDI-derived information as a feedback resource (Hamond et al., 2019; Hamond et al., 2020). None of this is new, in other words — neither pedal measurement nor pedal visualization is a novel idea by itself.

What stays genuinely difficult is interpreting the evidence correctly. A MIDI controller state is not an acoustic recording. An acoustic estimate of pedal depth is not a direct measurement of foot movement. A note that keeps sounding after key release isn't evidence the finger stayed on the key. A deep pedal depression isn't automatically the musically right choice. And conversely, a visually irregular pedal curve isn't necessarily a defect — expressive performance runs on individual and context-sensitive strategies that don't look tidy on a graph (Bernays & Traube, 2014; Lipke-Perry et al., 2022).

So this article asks a narrower question: how should digital sustain-pedal feedback separate what was measured from what might be pedagogically inferred? Through a structured applied review of piano pedagogy, performance science, acoustics, music information retrieval, and technology-mediated feedback, it builds a five-layer distinction — key action, pedal action, acoustic consequence, musical-interpretive context, pedagogical inference — and then proposes the Pedal Feedback Evidence Model (PFEM), an author-developed and unvalidated framework for designing and interpreting pedal feedback. The model doesn't judge pedaling quality, prescribe one universal technique, or claim learning benefits. Its whole purpose is to make evidence boundaries explicit before a digital representation quietly becomes a teaching claim.

3. Method

This is a structured applied literature-review design. It isn't trying to estimate an intervention effect, build a universal model of correct pedaling, or claim exhaustive coverage of every piece of piano-pedal literature. Instead it synthesizes heterogeneous evidence around one practical educational problem: the relationship between a digitally observable pedal-related signal and the pedagogical meaning people attach to that signal. This follows integrative review principles that let different forms of evidence sit together as long as the synthesis preserves each source's role and limitations, rather than treating unlike studies as if they measured the same thing (Whittemore & Knafl, 2005).

Searches ran, and were updated, on 18 August 2026. Targeted title, abstract, keyword, and citation-chaining searches covered scholarly web search plus primary publisher and index platforms — PubMed, SAGE, Frontiers, Intellect, the Audio Engineering Society, ISMIR proceedings, Zenodo, and university repositories. The official MIDI Association documentation supplied the MIDI 1.0 CC64 specification. Search families combined piano with sustain pedal, damper pedal, pedaling, or part-pedaling; pedal timing, pedal depth, hand-foot coordination, auditory feedback, attention, and individuality; acoustics, decay, sympathetic resonance, and damper-string interaction; MIDI, CC64, sensing, visualization, and automatic transcription; and piano pedagogy with visual, auditory, or technology-mediated feedback. Backward citation chaining from the most directly relevant works recovered historical and pedagogical prior art.

Sources fell into four evidence roles. Direct performance and pedagogy evidence covered empirical or scholarly work on pedal timing, techniques, teaching, attention, auditory feedback, or interpretive variability. Acoustic-mechanism evidence covered research on sustain, damping, decay, resonance, and part-pedaling. Digital-measurement evidence covered sensing, MIDI, pedal detection, pedal-depth estimation, visualization, and transcription studies. Pedagogical-bridge evidence covered technology-mediated feedback, piano timbre teaching, and the link between observable performance information and teacher-student interpretation. Commercial product descriptions, generic web tutorials, and non-piano uses of the word "pedal" were left out of the evidentiary synthesis entirely.

Emerging technical work got treated separately whenever peer-review status was incomplete or a conference version couldn't yet be fully reconciled with its open preprint — used to flag current computational directions, not to establish pedagogical effects. Research families were deduplicated conceptually: a conference prototype, its later journal article, and a subsequent detection model from the same technical program counted as related prior art, not three independent demonstrations of educational benefit.

The synthesis organized around a claim-evidence boundary. For each technical representation, five questions guided the analysis: What physical or acoustic event is actually being observed? How does it get represented digitally? Which interpretation does that representation directly support? What confounds or missing variables could change the meaning? And what additional observation or teacher action would be needed before making a pedagogical judgment? This structure is what generated the five-layer PFEM proposed later in the article.

This is not a systematic review or meta-analysis. A single reviewer conducted it, with OpenAI ChatGPT (GPT-5.6 Sol) supporting search-term expansion, evidence organization, drafting, and document preparation; none of its output counted as a scholarly source. The search is broad and deliberately adversarial, but still search-bounded — an equivalent framework may well exist under terminology this search didn't recover. A separate search audit records 15 search families, their evidence roles, and scope boundaries; a claim-evidence-contradiction ledger records 22 key claims or inference boundaries used to constrain the manuscript's wording. No new human-participant data, proprietary platform user data, product telemetry, or internal software tests get analyzed here as educational evidence.

4. Pedaling as Coordinated and Interpretive Action

Pedaling is a motor act and an interpretive practice at the same time. Repp's studies are especially useful here because they quantify the relationship between foot and hand instead of assuming that pedal markings translate directly into fixed motor timings. In a preliminary study of two pianists, pedal-release, pedal-onset, and pedal-change intervals varied with both local and global tempo, and the pattern differed substantially between the two performers (Repp, 1996). A follow-up study with ten pianists characterized hand-foot coordination as a highly practiced, perceptually guided form of complex motor behavior, and again found the timing relationship shifting with tempo rather than obeying one simple rule (Repp, 1997).

The pedagogical takeaway isn't that timing is arbitrary. It's that a numerical deviation from one reference timing can't, on its own, establish an error. The reference needs to be tied to a musical task, a chosen technique, a relevant comparison. Syncopated pedaling, for example, deliberately places the pedal change after a new key onset so continuity holds while the harmony refreshes; direct pedaling lines up key and pedal actions differently. Lipke-Perry et al. (2022) operationalized distinctions like these in a multiple-case study and found pedal use shifting depending on whether attention was directed internally, externally, or toward a metronome — with expertise and preparation tied to different patterns of consistency.

Historical and contemporary pedagogy back up this contextual reading. Heinlein (1929) explicitly investigated individual differences in damper-pedal performance and treated pedaling as part of interpretive individuality. Liu (2020), in a doctoral pedagogical study, organizes pedal instruction from basic press/release and depth through simultaneous, syncopated, and more creative techniques. Al Bakri (2025) identifies thirteen types or uses of the right pedal and documents real terminological discrepancies across piano schools and teaching traditions, emphasizing timing, release, speed, depth, duration, partial depression, and auditory control as dimensions that interact rather than operate independently.

These dimensions make a single correctness score genuinely problematic. A student's pedal action might be mechanically different from a teacher's model while still working musically; two expert pianists can reach related sonic goals through entirely different timing or depth strategies. Bernays and Traube (2014) offer a broader empirical parallel: expert pianists showed idiosyncratic patterns across articulation, touch, dynamics, and pedaling while still producing the requested timbral intentions. Fine-grained continuous pedal information revealed differences that coarser MIDI-like measures would have missed entirely. Performer individuality doesn't make feedback impossible — it changes what feedback is allowed to claim. A system can accurately describe a timing relation without having any authority to call that relation aesthetically wrong.

From Key Release to Acoustic Sustain

The gap between action and sound is clearest at the moment of key release. On an acoustic piano with no sustain, releasing a key returns the damper to the string and changes how the note decays. Even on a digital piano, Repp (1995) showed measurable decay in synthetic tones after key release — physical release and the sound actually disappearing aren't instantaneous equivalents. With the sustain pedal down, that gap widens considerably: the finger can release the key while the damper stays lifted and the sound continues on its own.

Lehtonen et al. (2007) measured several acoustic consequences of the sustain pedal directly. Beyond extending decay across parts of the keyboard, pedal use increased sympathetic vibration and changed how partials behaved — and the effect wasn't uniform across registers. That matters pedagogically because "how long a note sounds" isn't determined by key duration or pedal duration alone. Register, resonance, the specific instrument, overlapping harmonics, and whatever happens next all shape the acoustic result.

Part-pedaling destabilizes binary descriptions even further. Lehtonen et al. (2009) varied damper distance from the string and identified distinct phases — free vibration, damper-string interaction, final free decay. During the interaction phase, damping affected decay and timbre together. That's physical evidence that pedal depth changes acoustic consequence continuously, at least within a limited mechanical region — which means a display that labels the pedal simply ON or OFF is showing a controller decision, not the instrument's full acoustic state.

Auditory control is part of the same action-sound loop. Repp (1999) had pianists perform with and without auditory feedback and found that feedback deprivation produced generally subtle effects across most expressive parameters — but some pianists showed substantial changes specifically in pedaling. That doesn't prove every pedagogical intervention should privilege sound over other information, but it's a real caution against designing pedal feedback as though listening were optional. Visual data can surface timing and movement the ear doesn't isolate easily, but the acoustic consequence is still what pianists are actually regulating.

The literature on piano timbre points the same way. Li and Timmers (2021) found that timbre teaching in higher-education piano lessons involved constant interaction among concepts, bodily actions, sonic goals, teacher modeling, and student experience — timbre wasn't fixed information handed over intact from teacher to student. Pedaling sits inside this same ecology: a mechanical act gets its pedagogical meaning through the sound it helps create in a specific musical context, not in isolation.

5. What Digital Systems Actually Capture

Digital pedagogy gets more trustworthy once the representation is named precisely. "Pedal data" can mean several fundamentally different signals — a switch state, a continuous controller value, a measured physical displacement, an audio-derived estimate, a visual inference — and each carries a different evidentiary status.

5.1. MIDI CC64 and Binary Sustain State

The official MIDI 1.0 Control Change table defines controller 64 as Damper Pedal On/Off (Sustain): values at or below 63 read as off, values at or above 64 read as on (MIDI Association, n.d.). That gives a clear claim about the thresholded controller state — nothing more. It doesn't, on its own, establish that an intermediate controller value is a calibrated measurement of physical pedal depth, damper height, or acoustic damping. For legacy CC64 data, a digital system should describe what the controller reports rather than quietly relabeling that value as physical pedal position.

The 2026 MIDI Piano Profile changes the interoperability picture without erasing this boundary. The MIDI Association states that the Profile standardizes piano velocity and pedal curves and piano-specific controller behavior; its documented pedaling scope includes half-pedaling and continuous sustain alongside other piano pedals (MIDI Association, 2026). For conforming implementations, that gives compatible devices a negotiated behavioral contract and cuts down manufacturer-specific ambiguity at the MIDI layer. It still doesn't turn a transmitted controller value into a direct measurement of damper-string distance, guarantee identical acoustic consequences across every instrument and room, or establish that a given pedal trajectory is musically appropriate. PFEM therefore distinguishes legacy thresholded CC64, Profile-conformant controller behavior, calibrated physical sensing, and audio-derived estimation as separate provenance classes.

A binary CC64 trace is still pedagogically useful — it can show whether sustain engaged under the threshold rule, reveal broad relationships between pedal changes and note onsets, support replay and annotation. Its limitation matters just as much: once data get reduced to on/off state, the representation can't distinguish full pedaling from partial-depth behavior, and it can't tell the teacher what an acoustic piano or room actually sounded like. Profile-conformant continuous behavior preserves more controller information than this binary reduction does, but the correct pedagogical description still comes down to provenance — controller behavior is not the same thing as acoustic result, and neither is the same thing as musical quality.

5.2. Continuous Physical Pedal Measurements

Sensor systems can observe considerably more. The Piano Pedaller, built by Liang et al. (2017), used an optical sensor mounted inside the pedal mechanism to record pedal gesture alongside audio, classify techniques, and visualize results in a score-following environment. Liang et al. (2018) then reported measurement, recognition, and visualization of pedaling gestures and techniques in the Journal of the Audio Engineering Society. This is decisive prior art — measurement, classification, and visualization of sustain-pedal gestures all predate the present framework by years.

Continuous pedal-position data can describe onset, release, depth trajectory, press/hold/release segments, and repeated or partial motions. Bernays and Traube (2014) used similar high-resolution key and pedal position tracking to study expressive individuality. But physical depth is still only one layer: the same numerical depth won't necessarily produce identical acoustic results across instruments, because pedal mechanisms, damper geometry, regulation, and sound generation all differ from one piano to the next.

5.3. Audio-Derived Pedal Estimates

When no pedal sensor or MIDI stream exists, automatic systems can try inferring pedal use from audio instead. Liang et al. (2019) demonstrated transfer learning for sustain-pedal detection in acoustic piano recordings. Zhang et al. (2025), more recently, moved from binary detection toward high-resolution continuous pedal-depth estimation. Their result matters for two reasons: depth turns out to be estimable, but the study also reveals a real limitation — models tested in unseen room conditions weren't reliable, and reverberation introduced an overestimation bias. An audio-derived pedal curve, in other words, is an estimate shaped by the acoustic path, not a direct measurement of the foot.

That limitation carries real pedagogical weight. A teacher looking at an automatically inferred "80% pedal" value might read it as physical movement, when the model is actually responding to acoustic evidence shaped by room reverberation, instrument resonance, microphone position, and its own internal assumptions. A trustworthy interface should label a value like this as estimated and, where possible, expose confidence or uncertainty rather than rendering it with the visual authority of measured ground truth.

5.4. Key Action Versus Sounding Duration

Automatic piano transcription reveals another common collapse. The ATEPP project explicitly modified transcription so the modeled note offset reflects the pianist's key action rather than string damping time — because sustain can come from either the key or the pedal (Zhang et al., 2022). The 2024 MIREX Polyphonic Transcription task likewise separates raw note offsets from pedal-extended note durations in its evaluation conventions (MIREX, 2024). Wei et al. (2024) built sustain-pedal validation into streaming onset/offset decoding, again treating note offset and pedal state as things that need coordinating, not one single event.

The distinction is simple to state: KeyOff answers "when did the finger release the key?" A pedal-extended offset answers something different — "until when should this symbolic note count as sustained under a chosen representation rule?" Neither one is the same as the moment a human listener stops hearing acoustically meaningful energy. This is one of the most important boundaries for educational software to get right. If a system colors a note as lasting four seconds because of pedal extension, it must not tell the student the finger was held down for four seconds.

A very recent visual-transcription study reinforces this same boundary from another angle entirely. Kim et al. (2026), in an August 2026 arXiv version accepted to ISMIR 2026, state directly that sustain can make audio systems predict pedal-extended offsets rather than physical key release — and their work uses video to target physical key-off behavior separately, as its own signal. That's not pedagogical validation, but it strengthens the representational point: a system should say plainly whether an offset means key release, a pedal-extended symbolic duration, or an acoustically inferred endpoint.

5.5. Action- and Gesture-Level Analysis

Recent work pushes pedal analysis past individual frames. Zhang et al. (2026) evaluated continuous pedal-depth estimation using press, hold, and release segments alongside complete press-release gesture contours. This is computational work, not a pedagogical validation study, but it reinforces something important: the analysis unit needs to match the behavior actually in question. Frame-level error can punish small temporal shifts while missing a gesture's overall structural shape; binary F1 can hide half-pedaling and other continuous behavior entirely. For teaching, the lesson is methodological — feedback should choose a representation whose unit corresponds to the pedagogical question being asked.

6. From Performance Data to Pedagogical Feedback

Digital feedback has already found its way into piano teaching, in forms that combine auditory, visual, score-based, and MIDI-derived information. Hamond et al. (2019) reported an exploratory use of visual feedback for dynamics in higher-education piano learning. Hamond et al. (2020), working with three teacher-student pairs, found that feedback in a technology-enhanced studio touched on music, performance, and MIDI-related parameters, with engagement varying case by case. These studies support the pedagogical relevance of inspectable performance representations — but their exploratory design doesn't justify a general claim that digital feedback improves piano learning across the board.

The distinction between data and pedagogy matters especially for pedal, because the real teaching target is usually sonic rather than mechanical. A teacher can ask a student to change the pedal earlier, later, faster, more shallowly, or more deeply — but those instructions normally serve a musical purpose: clarity of harmony, continuity of line, resonance, color, articulation, phrasing, stylistic fit. A pedal trace earns its pedagogical value by helping teacher and student inspect the relationship among action, sound, and intention. It doesn't earn that value by replacing the relationship with a graph.

Li and Timmers (2021) offer a useful analogy from timbre teaching. Their observed lessons show sonic goals and embodied actions getting negotiated through interaction, not delivered as fixed objective rules. The same logic applies to sustain-pedal feedback: a visualization can become a shared object for discussion. The teacher might notice a release that consistently comes right before a harmonic change, compare that action against the audible blur it produces, ask the student to repeat the passage, and then judge whether the new action gets closer to the intended result. The visualization supplies evidence; the lesson supplies the interpretation.

A digitally generated statement, then, can sit at genuinely different epistemic levels. "CC64 crossed the on threshold at 12.42 s" directly describes a MIDI event. "The pedal was approximately 40% depressed" might be a direct sensor measurement if calibrated, or an estimate if inferred from audio — depends entirely on source. "The previous harmony continued into the next chord" is an acoustic or symbolic interpretation that needs additional evidence to support it. "The pedaling is too late" is a pedagogical judgment that needs a goal and a context to mean anything. And "the student does not understand legato pedaling" is a learner-level inference that no single pedal trace can establish. Keeping these levels separate is the whole design problem PFEM is built to address.

7. Pedal Feedback Evidence Model (PFEM)

The Pedal Feedback Evidence Model (PFEM) is proposed here as an author-developed, unvalidated framework for organizing sustain-pedal feedback. It doesn't classify students, score musical quality, or prescribe one canonical pedal technique. Its job is to make explicit the evidentiary jump from measured event to pedagogical claim — a jump that gets made silently far too often.

PFEM separates five layers. Layer 1, key action, covers the pianist's interaction with the keys: onset, release, hold, overlap, rearticulation. Layer 2, pedal action, covers foot-controlled pedal state, timing, depth, direction, gesture. Layer 3, acoustic consequence, covers decay, damping, sympathetic resonance, spectral change, and whatever sound persists after key or pedal events. Layer 4, musical-interpretive context, covers harmony, texture, articulation, phrase, style, tempo, instrument, room, score indications, performer intention. Layer 5, pedagogical inference, covers the teacher's interpretation and intervention — what to discuss, ask, demonstrate, compare, practice.

These layers are causally related in performance, but that doesn't make them interchangeable as evidence. Key action can contribute to acoustic duration; pedal action can alter damping; room acoustics can change what actually reaches the microphone; musical context determines whether a given overlap is desirable at all. A digital system might observe one layer directly and estimate another — but the interface needs to preserve that distinction rather than blur it.

PFEM asks every feedback item to answer five questions: What was measured? What representation resulted from that measurement? What claim does that representation directly warrant? What could confound that claim? And what verification or pedagogical action should follow before anyone acts on it? A sixth, optional field records provenance — MIDI, calibrated sensor, microphone audio, score annotation, teacher observation — letting a teacher tell direct measurement apart from algorithmic inference before drawing any conclusions from it.

Within the scholarly and technical sources this search identified, measurement, visualization, pedal-technique classification, binary detection, continuous-depth estimation, and technology-mediated piano feedback are all established prior art. No source combined key action, pedal action, acoustic consequence, musical-interpretive context, and an explicit pedagogical inference boundary in one teacher-facing evidence model. That's a search-bounded observation, not a claim of global priority or first-in-field status.

Table 1. Pedal Feedback Evidence Model (PFEM): proposed and unvalidated framework.

Evidence layer	Primary source	What can be represented	Directly warranted claim	Main confounds / missing evidence	Teacher verification
1. Key action	MIDI NoteOn/NoteOff; calibrated key sensor	Key onset/release, overlap, hold time	Finger/key behavior at measured interface	Pedal, acoustic decay, finger substitution, representation convention	Compare with pedal and audio before inferring sounding duration
2. Pedal action	Legacy CC64; Piano Profile-conformant controller; continuous pedal sensor; visual foot tracking	On/off state; timing; depth/trajectory; press-hold-release gesture	Pedal/controller behavior or estimated behavior, depending on source	Device calibration, mechanism, sensor mapping, thresholding	Label measured vs. estimated; inspect relation to keys and sound
3. Acoustic consequence	Microphone/audio features; instrument model	Decay, damping, resonance, spectral persistence	Observed/estimated sonic result	Room reverberation, mic position, instrument/register, overlapping notes	Listen and compare repetitions; avoid inferring foot motion from audio alone
4. Musical-interpretive context	Score, teacher/learner intention, repertoire and acoustic context	Harmony, articulation, phrase, texture, style, desired resonance	Why a pedal action may be appropriate in this passage	Competing interpretations, room/instrument differences, stylistic choices	State the musical goal before judging the trace
5. Pedagogical inference	Teacher synthesis across layers	Question, explanation, demonstration, comparison, practice task	A context-specific teaching decision	Expertise, learner history, incomplete evidence, one-trial variability	Use multiple evidence channels; avoid trait/ability claims from one trace

Note. PFEM is a proposed conceptual framework; it has not been validated as an assessment scale, diagnostic instrument, causal model, or predictor of learning.

Table 1. Pedal Feedback Evidence Model (PFEM): proposed and unvalidated framework.

Evidence layer	Primary source	What can be represented	Directly warranted claim	Main confounds / missing evidence	Teacher verification
1. Key action	MIDI NoteOn/NoteOff; calibrated key sensor	Key onset/release, overlap, hold time	Finger/key behavior at measured interface	Pedal, acoustic decay, finger substitution, representation convention	Compare with pedal and audio before inferring sounding duration
2. Pedal action	Legacy CC64; Piano Profile-conformant controller; continuous pedal sensor; visual foot tracking	On/off state; timing; depth/trajectory; press-hold-release gesture	Pedal/controller behavior or estimated behavior, depending on source	Device calibration, mechanism, sensor mapping, thresholding	Label measured vs. estimated; inspect relation to keys and sound
3. Acoustic consequence	Microphone/audio features; instrument model	Decay, damping, resonance, spectral persistence	Observed/estimated sonic result	Room reverberation, mic position, instrument/register, overlapping notes	Listen and compare repetitions; avoid inferring foot motion from audio alone
4. Musical-interpretive context	Score, teacher/learner intention, repertoire and acoustic context	Harmony, articulation, phrase, texture, style, desired resonance	Why a pedal action may be appropriate in this passage	Competing interpretations, room/instrument differences, stylistic choices	State the musical goal before judging the trace
5. Pedagogical inference	Teacher synthesis across layers	Question, explanation, demonstration, comparison, practice task	A context-specific teaching decision	Expertise, learner history, incomplete evidence, one-trial variability	Use multiple evidence channels; avoid trait/ability claims from one trace

Note. PFEM is a proposed conceptual framework; it has not been validated as an assessment scale, diagnostic instrument, causal model, or predictor of learning.

8. Design Principles for Digital Pedal Feedback

PFEM suggests several design principles for digital studios. The first is provenance before judgment. A display should identify whether a curve comes from legacy thresholded CC64, a Piano Profile-conformant controller path, a calibrated position sensor, audio inference, or computer vision. A smooth line simply labeled "pedal" hides a distinction that actually matters: one system might be displaying protocol/controller behavior while another is measuring physical displacement or estimating damper-related acoustic behavior — and those are not the same claim.

The second principle is keeping key and pedal events separate. Extending note durations according to pedal state can help with playback and symbolic representation, but educational interfaces should retain the original key release. A teacher may need to know whether continuity came from finger legato, finger overlap, pedal, or some combination — a single merged note bar destroys exactly that information.

Third, binary and continuous views shouldn't compete as rivals. A binary on/off representation works well for introducing basic syncopated pedal timing or quickly locating changes. Continuous depth becomes relevant once the lesson turns to half-pedaling, fluttering, gradual release, repeated shallow movements, or the shape of the pedal gesture itself. The interface should raise representational resolution when the teaching question calls for it — not on the assumption that more data is inherently better.

Fourth, acoustic feedback should stay available throughout. Repp (1999) and the acoustic studies reviewed above make it hard to justify a pedal-learning interface that pushes students to optimize a visual trace while ignoring sound. A better design lets the teacher align a pedal curve with replay, isolate a short passage, compare two attempts, and ask whether a measured change actually produced the clarity or resonance intended.

Fifth, uncertainty should be visible whenever a signal is inferred rather than measured. Zhang et al. (2025) show room reverberation biasing audio-based depth estimation. A system should avoid presenting an estimated value with the same visual status as calibrated sensor data — confidence bands, uncertainty labels, or plain categorical wording like "estimated pedal depth" would be more honest about what the evidence actually supports.

Sixth, feedback should favor questions and comparisons over categorical verdicts whenever musical appropriateness depends on context, which is most of the time. Instead of "Pedal error: 120 ms late," a system can report "pedal change occurred 120 ms after the new chord onset" and let the teacher compare the sounding result or a reference performance. That wording preserves a factual measurement while leaving the pedagogical decision exactly where the evidence can actually support it.

Seventh, digital systems should support repeated observation. Pedaling varies within a single performer across tempo, focus, and repetition (Repp, 1996, 1997; Lipke-Perry et al., 2022). One attempt is weak evidence for a stable learning problem. Trend displays, repeated-passage comparisons, and within-student baselines are probably more defensible than ranking a single trial against some universal template.

9. When Pedal Data Can Mislead

Getting real value from digital pedal feedback means preserving the cases where the data resists a simple interpretation. First: movement variability isn't automatically error. Repp's work shows systematic sensitivity to tempo and performer; Lipke-Perry et al. (2022) show attentional focus altering pedaling on its own. Bernays and Traube (2014) demonstrate idiosyncratic performance profiles involving pedaling as a matter of course. A template that penalizes any deviation from one model performance risks punishing legitimate individuality.

Second: more measurement resolution doesn't eliminate uncertainty — it just relocates it. Continuous pedal sensing reveals more than binary CC64 does, but acoustic consequence still depends on mechanism and instrument. Audio estimation observes the consequence more directly but becomes vulnerable to room acoustics and can't guarantee recovering the underlying physical gesture. The two modalities are answering different questions, not competing answers to the same one.

Third: score markings are incomplete ground truth. Pedal indications can specify intended changes or effects, but they often leave depth, release speed, local adjustment, and room-dependent decisions entirely unspecified. Al Bakri (2025) documents both the terminological variation and the real weight of interpretive judgment involved. A digital system shouldn't treat a printed Ped. symbol as a complete motion trajectory.

Fourth: acoustic cleanliness isn't a universal optimization target. Some repertoire and interpretive goals actively call for resonance, blending, color, or controlled harmonic carryover that would look "messy" under a simplistic overlap metric. Conversely, a mechanically precise pedal change can still sound unsatisfying on a particular instrument or in a reverberant room. This is exactly why PFEM places musical-interpretive context between acoustic consequence and pedagogical inference — not adjacent to them.

Fifth: visual feedback can redirect attention in ways that change the very behavior it's trying to measure. Technology-mediated feedback research in piano suggests real value alongside variable engagement (Hamond et al., 2020). Lipke-Perry et al.'s (2022) pedal study is a specific warning here: shifting attentional focus can itself alter motor output. A system that asks students to watch the pedal curve continuously might change the behavior it's meant to observe. Whether feedback arrives during performance or after a take should be treated as a research question, not an assumed design choice.

Finally: data shouldn't get escalated into learner-level labels. A late release in one passage doesn't establish that a student "doesn't understand pedaling," and a visually consistent curve doesn't prove aural control, stylistic knowledge, or transfer to another piano. Claims like these aren't supported by the current evidence, and they sit explicitly outside what PFEM is built to do.

10. Applications for Piano Teaching

A beginner lesson can use PFEM at low resolution. The teacher might show a binary pedal trace aligned with key onsets to illustrate the difference between simultaneous and syncopated change. The instructional question is purely temporal: did the pedal release before, with, or after the new harmony? Teacher and student listen to the effect together. There's no need for a continuous depth graph if depth isn't yet the target.

At an intermediate level, the same passage can come back with separate key-release and pedal traces shown side by side. This helps distinguish finger legato from pedal legato — a note bar that stays visually extended because of sustain shouldn't obscure when the finger actually released. The teacher can ask the student to produce the same continuity using different combinations of finger overlap and pedal, making the relationship between motor strategy and sonic result concrete and visible.

When half-pedaling or gradual release becomes the target, a Piano Profile-conformant controller path, a calibrated continuous sensor, or another device whose intermediate values have been validated for the intended measurement becomes appropriate. The teacher can compare depth trajectories across repetitions, but the goal still needs to be stated musically — clearing the bass without eliminating all resonance, say. The trace becomes evidence about the action pursuing that goal, not a definition of the goal itself.

For advanced repertoire, teachers can use synchronized audio, score position, key events, and pedal trajectory together to work through a short passage. A blurred harmonic change can be inspected layer by layer: was the key still held? Was the pedal release late? Was the release too shallow to re-engage the damper properly? Did room reverberation create an impression of carryover even after the mechanical change happened? Was the overlap intentional in the first place? PFEM turns these into distinguishable questions instead of one vague impression.

The framework can also support teacher education directly. Students training to teach can be given several representations of the same performance and asked to sort each statement into direct observation, algorithmic estimate, musical interpretation, or pedagogical inference. Exercises like this train evidence literacy — the ability to say not just what a graph appears to show, but what the measurement actually warrants saying.

For designers, the practical recommendation is to make uncertainty and layer-switching part of the interface itself. A user should be able to toggle among key action, pedal action, audio consequence, and score context, rather than receiving one composite "pedal score." That makes the software look less decisive, but it's more honest about what the evidence supports — and more useful for a teacher trying to interpret it.

The National Association for Music Education (NAfME, 2022) emphasizes pathways for practitioners to understand and apply research and effective practice. PFEM translates that same professional-learning priority into an evidence-literacy routine: identify the measurement source, inspect the audible result, and save pedagogical judgment for evidence that can actually support it.

11. Limitations and Research Agenda

This review has real limits. It's a structured applied review conducted by a single reviewer, not a systematic review or meta-analysis. Searches covered multiple primary publisher and index platforms and included backward citation chaining, but no claim is made here that every historical pedal treatise, national pedagogical tradition, engineering paper, or unpublished study got retrieved. The novelty observation about PFEM is therefore search-bounded, not a claim of priority.

The evidence base is heterogeneous by nature. Acoustic studies measure string and damper behavior; performance studies measure timing and individual variation; computational studies measure detection or estimation accuracy; pedagogical studies examine teacher-student interaction or technology-mediated feedback. These bodies of evidence can't be pooled as though they estimate one outcome. Their real value here is triangulation — they reveal different boundaries along the chain from action to sound to interpretation, and each one only sees part of it.

PFEM itself remains unvalidated. Its categories haven't been tested for inter-rater reliability, teacher usability, instructional efficiency, or learning effect. The first empirical step should be descriptive validation with experienced piano teachers: do they actually recognize the five layers as distinct and useful? Which pedal-learning tasks call for which evidence sources? Where do teachers disagree with each other?

A second research program should examine measurement validity directly. Studies could compare legacy thresholded CC64, Piano Profile-conformant continuous controller data, calibrated pedal sensing, optical/video foot tracking, and audio-based depth estimation on the same acoustic piano — not to crown one representation universally superior, but to quantify when each one agrees with physical pedal motion and when protocol mapping, mechanism, room, or instrument factors pull them apart.

A third program should study pedagogical interpretation itself. Teachers could review identical performances under different feedback conditions — audio only, binary pedal trace, continuous trace, separated key/pedal events, full multimodal representation — while researchers track what teachers notice, what feedback they give, and whether more data actually improves agreement or just adds cognitive load. Because visual attention can alter motor behavior on its own, delayed versus real-time feedback should get compared too.

Finally, learning claims would need longitudinal studies before anyone can responsibly make them. The useful question isn't simply whether students can imitate a target pedal curve — it's whether they develop more reliable auditory control, adapt across instruments and rooms, choose pedal strategies appropriate to musical context, and can explain the relationship between action and sonic result themselves. Those outcomes sit deliberately outside what this review can establish.

12. Conclusion

The sustain pedal makes a poor candidate for simplistic digital scoring, because its meaning crosses several domains at once. The finger releases a key. The foot changes a pedal. Dampers move. Strings keep vibrating or stop. Room acoustics reshape the result. And somewhere in all of that, a musician decides whether the sonority actually fits a phrase, a harmony, a texture, a style, an intention. A digital representation can capture parts of that chain with real precision while still falling far short of justifying the final pedagogical judgment.

The literature reviewed here supports several firm distinctions. Pedal timing varies with performer and tempo. Part-pedaling changes acoustic behavior in ways binary states simply can't describe. Legacy MIDI 1.0 CC64 provides useful thresholded on/off semantics, while the 2026 MIDI Piano Profile standardizes richer piano-specific pedal behavior — including half-pedaling and continuous sustain — for conforming implementations. Neither protocol layer, by itself, measures acoustic damper-string interaction or establishes musical correctness. Sensor-based systems can measure and visualize gesture more directly; audio-based systems estimate pedal behavior through an acoustic path that room reverberation can bias. Automatic transcription still has to decide whether an offset represents physical key release or pedal-extended sustain. Technology-mediated feedback can put performance parameters on the table for discussion, but availability alone doesn't establish musical correctness, and it doesn't establish learning either.

The proposed Pedal Feedback Evidence Model responds to all of this by separating key action, pedal action, acoustic consequence, musical-interpretive context, and pedagogical inference. Its contribution is deliberately modest: a vocabulary for stopping one evidence layer from getting silently substituted for another. PFEM doesn't tell a teacher what good pedaling has to sound like. It asks a prior question that digital studios increasingly need an answer to — what exactly did the system observe, and how far can that observation legitimately travel before a teacher has to listen, interpret, ask, and decide for themselves?