01 / REPEATED NOTES
Check attacks, not just note count
Our real single-note guitar MP3 produced three browser-downloaded events, all pitch 58, against seven reference events spanning pitches 57, 58 and 60. Evidence.
Only one of those three events matched equal reference pitch and onset within 150 milliseconds, leaving two unmatched outputs and six missed reference events. Evidence.
Cleaner, Balanced and More notes returned one, three and two events respectively, while their exact-pitch onset match counts were zero, one and zero. Evidence.
The longest More notes event lasted about 1.207292 seconds, compared with about 0.929167 seconds in Balanced. Evidence.
More notes admitted 46 RMS-qualified analysis windows versus 37 in Balanced, despite exporting fewer separate events. Evidence.
Against the uncompressed control, this MP3 kept the three-note pitch sequence but moved at least one exported onset by 46.875 milliseconds. Evidence.
The last note’s velocity also changed from 80 in the WAV control to 79 in the MP3 output. Evidence.
Recorded MP3 · Actual browser MIDI · Pitch/time ledger · Piano Roll
02 / CHORD SCOPE
More events are not more voices
The real guitar-chord MP3 downloaded from the browser contains four events with pitches 39, 43, 46 and 91, compared with seventeen reference events. Evidence.
None of those four output pitches occurs in the reference pitch set, so the baseline has zero exact-pitch onset matches. Evidence.
Encoding this crop to MP3 reduced the baseline from five WAV-control events to four, with events above the reference maximum decreasing from two to one. Evidence.
On that same MP3, Cleaner, Balanced and More notes exported one, four and seven events respectively. Evidence.
More notes gained two exact-pitch onset matches, but five output events remained unmatched and fifteen reference events were missed. Evidence.
All three MP3 settings reached only one simultaneous output note, where this crop’s reference reaches four held notes. Evidence.
Treat octave-related candidates, simultaneous-note coverage and unsupported high events as separate checks; none of them alone proves complete chord recovery. Evidence.
Recorded MP3 · Actual browser MIDI · Pitch/time ledger · More notes Piano Roll
03 / STEREO CONTROL
Inspect the channel mix first
From one decoded stereo piano MP3, averaging both channels produced two events, while the left-only and right-only diagnostic inputs produced six and three. Evidence.
Their pitch sets were 40 and 62 for the mean, 33, 40, 62 and 74 for the left channel, and 40, 52 and 62 for the right. Evidence.
The mean and right-channel diagnostics had zero exact-pitch onset matches, while the left-channel diagnostic had one. Evidence.
That lone left-channel match begins about 142.292 milliseconds late, leaving only about 7.708 milliseconds inside the chosen tolerance. Evidence.
Only about 28.43% of that output interval overlaps its matching hardware key-held interval. Evidence.
The mean, left and right diagnostics passed 27, 51 and 25 analysis windows through the same RMS threshold. Evidence.
Their whole-crop RMS values were about 0.008434, 0.011256 and 0.008443 respectively, on a full-scale amplitude basis. Evidence.
The piano WAV control and MP3 mean produced identical stored pitches and times, but the second note’s velocity changed from 55 to 54. Evidence.
Left-only and right-only are offline source diagnostics, not extra conversion modes offered by the page. Evidence.
Recorded stereo MP3 · Actual browser MIDI · Left diagnostic ledger · Left diagnostic Piano Roll
FILES BEFORE CLAIMS
Reproduce, then decide what to edit
These MP3s were encoded at 192 kb/s from archived human microphone recordings using one fixed FFmpeg build, without normalization or synthesized replacement instruments. Evidence.
The numerical controls decode to signed-16-bit PCM at 44,100 Hz and use the unchanged production detector and MIDI writer in a local JavaScript harness. Evidence.
Three actual browser downloads are byte-identical to their corresponding Balanced harness outputs in this run. Evidence.
The independent checker matches equal integer pitch to the nearest unused reference onset within 150 milliseconds, without accepting a note by duration or octave equivalence. Evidence.
GuitarSet references are partly automated, while the piano reference uses hardware key events and a previously disclosed −0.63375-second clock offset with timing uncertainty. Evidence.
These short examples change performance, room, dynamics and annotation as well as instrument, and cannot rank converters or isolate every acoustic cause. Evidence.
We did not conduct a listening study, import into a DAW, benchmark competing services or reopen the sealed guitar lag diagnosis. Evidence.
Reproduction contract · References · Parsed results · Browser/control comparison · Fresh-directory hashes · Download SHA256 · November 9 baseline ledger · November 9 baseline count
python generate.py --ffmpeg ffmpeg
node convert.mjs
python analyze.py
python compare.py
04 / KEEP THE WANTED ATTACKS
More matched attacks can still mean more cleanup
These editing checks use four adjacent five-second excerpts, labeled A through D, of one human GuitarSet recording, a fixed Balanced setting and inherited reference labels, with slowed MIDI mapped to the original timeline before comparison. Evidence.
The half-tempo WAVs lasted about 9.975–9.980 seconds instead of exactly ten, whereas the half-rate files lasted exactly ten, so the two operations did not even produce identical nominally restored clip lengths. Evidence.
For B, half-tempo preprocessing increased onset matches from two to five, but also raised unmatched output events from seven to ten as total events grew from nine to fifteen. Evidence.
For C, going from 0.75 to 0.50 speed added two events, from eleven to thirteen, without adding any onset matches beyond eight, so unmatched events rose from three to five. Evidence.
D’s neighboring reference pitch-60 attacks numbered 5 and 6 had no accepted baseline counterpart under our rule, while the half-tempo draft matched both. Evidence.
Recorded B MP3 · Original-speed MIDI · Half-tempo restored MIDI · C Piano Roll · D attack ledger
05 / CHECK THE CORRECTED NOTES
A clock or octave correction is only one check
For A’s pitch-62 attack, changing half-tempo preprocessing from 0.75 to 0.50 speed moved onset error from about −59.336 to +45.091 milliseconds, yet duration error worsened from about −34.494 to −115.744 milliseconds relative to the inherited interval. Evidence.
After time and octave correction, A’s half-rate and half-tempo drafts both had nine events and four matches, yet the half-rate draft contained three pitch-103 events absent from A’s reference labels and from the half-tempo draft. Evidence.
For C, corrected half-rate playback matched pitches 62 and 63 that half-tempo preprocessing missed, increasing matches from eight to ten while reducing unmatched events from five to one. Evidence.
B’s newly matched pitch-62 event in the corrected half-rate draft began only about 6.202 milliseconds late but was about 847.522 milliseconds shorter than its inherited reference interval. Evidence.
For A’s shared pitch-62 reference attack, corrected half-rate playback had about −0.119 milliseconds of duration error versus about −115.744 milliseconds for half-tempo preprocessing, despite both drafts sharing the same four-match total. Evidence.
Recorded A MP3 · Rate-corrected A MIDI · Tempo-corrected A MIDI · C corrected Piano Roll · B onset and ending ledger
06 / COMPARE THE ACCENTS
Restoring time does not restore every velocity
In B’s two shared matched attacks, pitches 65 and 70 had velocities 59 and 57 in the baseline but 59 and 60 after half-tempo preprocessing, reversing their stored velocity order even after time was restored. Evidence.
Stored MIDI velocity is an event value rather than a direct acoustic loudness measurement, so inspect it separately from playback volume. Evidence.
Keep the untouched output beside the restored draft and compare the wanted notes before editing their velocities. Evidence.
Baseline velocities · Slowed velocities · Restored B MIDI · Method, source and limits
Current whole-homepage ledger · Internal sentence audit
Choose the source
Use the MP3 to MIDI converter workflow for the one-note-at-a-time prerequisite and export checks. Evidence.
Check the format
The change MP3 to MIDI test separates an encoding change from an assumed timing repair. Evidence.
Check missing notes
The convert an MP3 file to MIDI test examines a deliberately quiet note under the available settings. Evidence.