M4A to MIDI:
when a WAV copy helps

To convert M4A to MIDI, use Local note transcription for a short recording of one instrument, then check and edit the downloaded notes; the tested AAC and sixteen-bit ALAC examples now complete locally, while an unsupported codec may still need a local WAV copy. Evidence

Open the converter · See the twelve attempts · Download the comparison files

Earlier original-detector baseline — retained unchanged

1. The same extension can reach different stages

Our controlled files put AAC or Apple Lossless audio in an M4A container and make a local PCM16 WAV counterpart, so the comparison changes the decoding path without claiming an authentic iPhone recording. Evidence

All three ALAC M4A inputs stopped with the browser’s decode error, whereas their WAV counterparts reached note detection, identifying a codec-entry failure separately from the notes found afterward. Evidence

Actual ALAC error · Same-source WAV result · Twelve actual UI records

2. Real guitar: a usable file still needs an ending check

The guitar’s compatible paths each returned six events against sixteen scored attacks, with one strict match, fifteen missed references and five outputs not matched as new attacks. Evidence

The accepted pitch-62 attack arrived 24.06 milliseconds early under the unchanged fifty-millisecond rule, giving one concrete timing check instead of a general accuracy promise. Evidence

Its stored duration was 81.18 milliseconds shorter than the reference duration, so acceptance of that attack did not establish the correct release. Evidence

The first saved pitch-58 event lasted 1.068 seconds from the crop boundary, while the same-pitch carried-in reference ended after 0.106 seconds, exposing a held-note length discrepancy outside the new-attack score. Evidence

The saved guitar events ended at 4.086 seconds, before the two annotated pitch-53 attacks at 4.356 and 4.598 seconds, so a file that downloads can still omit the phrase’s tail. Evidence

Carried-in notes are retained separately from newly annotated attacks, and an unmatched event under this rule is not automatically a fabricated sound. Evidence

3. First piano excerpt: decoded audio without a download

The first piano window contains twenty-five scored key attacks, but its AAC M4A and both WAV paths reported no stable melody notes and offered no MIDI download. Evidence

Its ALAC M4A stopped earlier at decoding, so making the WAV copy cleared that entry gate without recovering a note in this example. Evidence

The webpage used the same generic decode-failure heading for this no-note result, but the detailed message and audio duration distinguished it from the ALAC decoder error. Evidence

Actual detail: No stable melody notes were detected. · ALAC detail: Unable to decode audio data

4. Second piano excerpt: one low event is not the performance

The second piano window has twenty-four scored attacks, yet its compatible paths each saved one pitch-38 event with zero strict matches, twenty-four missed attacks and one unmatched output. Evidence

That event began at 4.644 seconds and lasted 186.46 milliseconds in the downloaded MIDI, giving a specific location to inspect before editing. Evidence

A separate fifty-millisecond pitch-class check finds references both one and three octaves higher, so the low event has two register candidates rather than a uniquely diagnosed octave error or a correct-pitch credit. Evidence

Three already-held piano keys are retained as crop context rather than counted as new attacks, and hardware key release is not the same measurement as audible pedal decay. Evidence

5. Twelve actual browser attempts

For each nonempty clip, the actual AAC M4A, AAC WAV and ALAC WAV downloads were byte-identical, so WAV preparation changed neither the accepted nor rejected note events in these saved results. Evidence

All six actual downloads also matched their corresponding offline MIDI bytes, while the three empty offline MIDI files remain offline artifacts rather than invented browser downloads. Evidence

Reference-order nearest-unused matching requires the exact integer pitch and an onset within fifty milliseconds, without a fitted time shift, octave credit or a relaxed posthoc window. Evidence

The AAC decoder produced 138 extra tail samples in this fixed FFmpeg path, but that input difference must not be turned into an unmeasured claim about a global Chrome or iPhone onset shift. Evidence

6. Download the inputs, notes and Piano Rolls

guitar15

Original crop WAV · Controlled AAC M4A · AAC-decoded WAV · Controlled ALAC M4A · ALAC-decoded WAV · Reference attacks · Offline output notes

piano36

Original crop WAV · Controlled AAC M4A · AAC-decoded WAV · Controlled ALAC M4A · ALAC-decoded WAV · Reference attacks · Offline output notes

piano61

Original crop WAV · Controlled AAC M4A · AAC-decoded WAV · Controlled ALAC M4A · ALAC-decoded WAV · Reference attacks · Offline output notes

The Piano Rolls are generated plots rather than DAW screenshots, and the empty piano plot represents the independently scored offline result because the webpage did not export a file for that case. Evidence

Fresh regeneration matched eighty-five common inputs, MIDI files, scores, plots and metadata hashes, validating the offline pipeline separately from the twelve native browser observations. Evidence

Frozen protocol · Input transforms and SHA · Actual native records · Independent result arithmetic · Original Windows input generator · Unchanged detector runner · Independent MIDI reader · Note scorer and plots · Recheck article numbers · Fresh-run hashes · Dated source comparison · Internal sentence ledger · Internal full-body audit · Evidence SHA256

7. Scope and local processing

The batch reuses licensed GuitarSet and Vienna4x22 human performances with declared crops and controlled exports, so it is not three independent performers, natural phone-capture validation or a mobile Safari test. Evidence

Guitar string-label uncertainty, piano timing calibration, carried-in notes and acoustic sustain limit what these annotation-relative counts establish, and no overall accuracy percentage is reported. Evidence

During the observed last WAV selection and one-event MIDI generation, Chrome’s unfiltered Network list remained empty, a scoped local-conversion check that does not cover initial assets, analytics, extensions or other files. Evidence

Network before selection · Generated result and network list · Unchanged client source · Licenses and modifications

8. Choose the next repair from the failure stage

Keep a local WAV fallback for a genuine decoder failure, then inspect the saved pitches, attacks and endings; an empty or incomplete single-pitch draft calls for a clearer part or a suitable transcription method rather than another filename change. Evidence

Audio-to-MIDI measured scope · Guitar checks · Piano checks · WAV decoder checks