MP4 to MIDI: extract the audio first.
Note Foundry does not accept MP4 files directly. Extract a mono MP3 or WAV track from the video, then convert that audio in your browser. In a reproducible test, the extracted AAC track recovered the same four MIDI pitches as its WAV baseline; a video with no audio produced no conversion input.
What the tested workflow does
We generated a four-note melody, packaged it into an AAC MP4, extracted its first audio stream with FFmpeg, and ran the extracted PCM through the current Note Foundry detector. The converter itself still receives MP3 or WAV, not MP4.
- Extract one audio stream from the MP4.
- Convert the extracted file and download the MIDI.
- Compare pitch and timing against the source before editing.
MP4 extraction results
The source melody used MIDI pitches 62, 65, 69 and 74. The WAV baseline and the AAC-in-MP4 extraction each returned those four pitches. The extracted notes began about 70 ms earlier than the generated event boundaries in both runs, so the container did not make timing exact.
| Input | Result | Evidence |
|---|---|---|
| WAV baseline | 4 notes: 62, 65, 69, 74 | WAV · MIDI |
| MP4 audio extracted to WAV | 4 notes: 62, 65, 69, 74 | MP4 · extracted WAV · MIDI |
| MP4 with first silent track selected | No clear notes | silent extraction |
| Video-only MP4 | FFmpeg reported no audio stream | video-only file |
Reproduce the test
Download run-mp4.cjs, the tested app.js, and the fixture files. The run uses FFmpeg to select an audio stream, resample to 11,025 Hz mono PCM, then evaluates the exact detector and MIDI writer. Full results, commands, source hash and the no-audio error are in results.json.
This is a synthetic isolated-melody test, not a real-video accuracy score. Browser decoding, noisy speech, chords and multi-track mixing need separate checks.
MP4 to MIDI questions
Can I upload an MP4 directly here?
No. Extract MP3 or WAV first, then use the converter.
What if the MP4 has multiple audio tracks?
Select the intended track during extraction. The test showed that selecting a silent first track returned no notes while the melody track worked.
Does extracting audio preserve the melody?
In this controlled AAC test, the four pitches were preserved. Timing still shifted, so inspect the MIDI against the source.