MP3 to MIDI Converter
Drop an audio file in and get a standard .mid file back. Vocal, piano and instrument modes run locally, and HQ Full Song can use your GPU to split the mix into six stems before writing multi-track MIDI.
Vocal uses GAME, piano uses Transkun V2, and guitar/bass/other instruments use Basic Pitch. Full Song can use Fast mixed-audio transcription or HQ Local six-stem WebGPU separation.
HQ Local becomes available when Full song is selected. All active conversion modes keep the source audio on your device.
Drop an audio or video file here
or click to choose — MP3, WAV, M4A, FLAC, OGG, AIFF, MP4, MOV
Ready · the selected model loads only when you convert
Compare the source with the detected notes directly. MIDI stores note events, not the original singer, instrument tone, mix or effects, so the preview deliberately uses a neutral synthesized sound.
Clean-up and tempo
This writes a General MIDI program into the downloaded file. DAWs can still replace it with any instrument.
Advanced
Piano roll
0:000:00
Supported formats
Decoding uses your browser's own audio decoders, so the same file that plays in your browser will convert here.
Audio
- MP3 .mp3
- WAV .wav
- M4A / AAC .m4a .aac
- FLAC .flac
- AIFF / AIF .aiff .aif
- OGG / OGA .ogg .oga
- Opus .opus
- WebM .webm .weba
Video (audio track is read, video is ignored)
- MP4 .mp4
- M4V .m4v
- MOV .mov
Browser transcription prepares the signal for the selected engine: GAME and Transkun use 44,100 Hz; Basic Pitch uses 22,050 Hz. HQ Local separates six 44.1 kHz stems with WebGPU before transcription.
How to convert MP3 to MIDI
- Drop your file in. Any of the formats above. Processing time depends on the selected engine and your device; HQ Local is much heavier than Fast mode.
- Wait for the piano roll. The page shows you exactly what it found — how many notes, over what range, where they sit in time.
- Download the .mid file. It is a standard MIDI file. Open it in Ableton, FL Studio, Logic, GarageBand, MuseScore or any editor and change the notes, the instrument, or the key.
This is transcription, not format conversion. Choose Vocal for singing, Piano for solo piano, or the matching instrument mode for isolated sources. Full Song Fast transcribes the mix directly; HQ Local uses your GPU to separate six stems before local multi-track transcription.
Frequently asked questions
Is this audio to MIDI converter free?
Yes. No account, no email, no credit card, no daily limit. The model runs on your own machine, so there is no server cost to pass on to you.
Is my file uploaded anywhere?
No active conversion mode uploads your source audio. Fast modes run in the browser, and HQ Local uses WebGPU on your device for stem separation before local transcription. Model files are downloaded when needed, but your song stays on your device.
Can I convert WAV to MIDI as well as MP3?
Yes — this is a WAV to MIDI converter, an M4A to MIDI converter, a FLAC to MIDI converter and an OGG to MIDI converter in the same box. If your browser can play the file, this page can read it. WAV and FLAC are uncompressed, so they tend to give slightly cleaner note detection than a low-bitrate MP3 of the same recording.
What comes out the other end?
A standard MIDI file (.mid), format 0, at 480 ticks per quarter note, carrying a tempo measured from the audio. Vocal mode writes the strongest melodic line; instrument mode can retain overlapping notes and chords. Import it into any DAW or notation app.
How accurate is the conversion?
It depends on the source, and this page does not invent a number for you. What it does instead is show you the result before you download it: the note count, the pitch range and the piano roll. If the roll looks wrong for your file, you know immediately instead of finding out in your DAW.
Does it work on a full song with several instruments?
Yes, with two choices. Fast mode transcribes the mastered mix directly. HQ Local uses WebGPU to separate six stems — Vocals, Piano, Guitar, Bass, Drums and Other — then transcribes them independently and writes a Format 1 multi-track MIDI. The drum track currently uses local onset/frequency classification after separation.
Is there a length or size limit?
No fixed limit is imposed by this page. Practical limits come from your device: very long files take proportionally longer and use more memory. A few minutes is comfortable on a laptop or a phone.
Does it work on a phone?
Yes. The page is built mobile-first — the drop zone, the buttons and the piano roll all fit a phone screen. Converting a long file on a phone will be slower than on a laptop.
What tempo is written into the MIDI file?
The tempo is measured from the note onsets and written into the file, and the note positions are computed from that same tempo — so the notes still land at their true times in the recording while the bar lines line up with the music. When no steady pulse can be found the file falls back to 120 BPM, and the panel tells you that happened. You can switch the measurement off, or set your own tempo in the DAW afterwards.
Which model is doing the transcription?
The site routes by source type. Vocal / humming uses GAME, piano uses Transkun V2, and guitar / bass / other instruments use Basic Pitch. HQ Local first uses BS-RoFormer WebGPU to split six stems, then routes each pitched stem to its matching engine; the separated drum stem is converted to GM drum events locally.