Audio to MIDI Converter

Turn a recording into editable MIDI notes with local, on-device transcription—free, private, and without an account.

Choose an audio file

Your recording and transcription stay in this browser. Nothing is uploaded.

MP3, WAV, OGG, FLAC, M4A, or AAC. Maximum 25 MB and five minutes.

Drop an audio file here

or choose one from your device

.mp3, .wav, .ogg, .flac, .m4a, .aac · 25 MB · 5 minutes

Choose an audio file to begin.

How to convert audio to MIDI

  1. 1. Choose a recording. Select or drop an MP3, WAV, OGG, FLAC, M4A, or AAC file. Your browser checks its type, size, and duration locally.
  2. 2. Set the sensitivity. Balanced suits most recordings. Clean reduces weak artifacts, while Detailed keeps more quiet and short notes.
  3. 3. Transcribe and download. The on-device model listens for pitched notes and builds a Standard MIDI file you can correct in a DAW or notation editor.

What note transcription can recover

The model estimates note pitch, start time, duration, and confidence-derived velocity. Clear isolated piano, guitar, bass, and other pitched-instrument recordings usually give a more useful starting point than a mastered song with several overlapping parts.

Audio does not carry a clean copy of the original MIDI instructions. The export cannot reliably recover instrument assignments, tempo changes, meter, key, pitch bends, lyrics, drums, effects, or separate performers. Treat the result as an editable draft, not an exact score.

Clean, Balanced, or Detailed

Sensitivity controls the tradeoff between missed quiet notes and extra noise artifacts. Start with Balanced. Move to Clean for noisy rooms, ringing effects, or too many weak notes. Try Detailed when a clean solo recording loses short or quiet notes. Changing sensitivity re-runs detection without changing your original file.

Formats, privacy, and browser limits

Files may use an .mp3, .wav, .ogg, .flac, .m4a, or .aac extension, but codecs vary across browsers and operating systems. If one browser cannot decode a valid file, converting it to ordinary PCM WAV can improve compatibility.

The 25 MB and five-minute limits bound decoding, resampling, and model memory. The recording and outputs remain on your device. Once you press Convert, the page loads Songparts's own worker and pinned model assets; there is no audio upload or third-party model request.

Plan on a correction pass

Open the downloaded MIDI in a DAW or notation editor, listen against the original, remove false notes, restore sustained notes, quantize only where useful, and assign instruments. The fixed 120 BPM export preserves absolute detected timing, so changing the project tempo without time-stretching the notes will change playback alignment.

If transcription fails or finds no notes

Use a current desktop browser, try a short PCM WAV excerpt, close memory-heavy tabs, and keep the pitched instrument prominent. Cancel is available during preparation and detection. A stalled worker stops with a retryable message; no-notes results can often improve with Detailed sensitivity or cleaner source audio.

FAQ

Is this Audio to MIDI converter free?
Yes. You can transcribe and download a MIDI without an account or payment.
Is my audio uploaded?
No. The recording, decoded samples, detected notes, and generated MIDI stay in your browser. After you press Convert, the browser loads Songparts's own transcription worker and model files; it does not send your audio anywhere.
Which audio formats can I convert?
You can choose MP3, WAV, OGG, FLAC, M4A, or AAC files. Actual decoding support depends on your browser, operating system, and the codec stored inside the file.
Which model detects the notes?
The tool runs Spotify's open-source Basic Pitch model locally in a dedicated browser worker. Songparts serves a pinned local copy of the model rather than fetching it from a third-party service.
How accurate is Audio to MIDI conversion?
It is approximate. Clear, isolated instruments with distinct notes usually give the cleanest starting point. Room noise, effects, expressive pitch, overlapping harmonics, and dense arrangements can create missed or extra notes.
Can it separate vocals, drums, or a full mix?
No. This tool detects pitched note events; it does not separate stems, recognize lyrics, transcribe drums, or identify chords as named symbols. Vocals and dense mixes may produce rough note guesses that need substantial editing.
Does the MIDI preserve tempo, instruments, and pitch bends?
No. Version one exports one neutral piano track at a fixed 120 BPM while preserving detected notes at their absolute times and durations. It does not recover the source tempo map, instruments, channels, meter, key, lyrics, or pitch-bend automation.
How long can the recording be, and how long does transcription take?
The input limit is 25 MB and five minutes. Processing time depends on duration, device speed, browser, and note density; current desktop browsers work best for longer recordings.
Which sensitivity should I choose?
Balanced is a practical default. Clean favors stronger notes and fewer noise artifacts. Detailed keeps more quiet and short notes, but may also keep more unwanted notes.
Which browsers work best?
Use a current desktop browser with Web Workers, Web Audio, typed arrays, and WebGL or a capable CPU. Mobile and memory-constrained devices can struggle with long recordings, and codec support differs by browser.
What should I do if transcription stalls, fails, or finds no notes?
Cancel and retry, choose a shorter excerpt, use a current desktop browser, and close memory-heavy tabs. For no notes, try Detailed sensitivity or a cleaner recording with a more prominent pitched instrument.
Can I edit the detected notes here?
Not in this version. Download the MIDI and correct timing, pitches, durations, and unwanted notes in a DAW or notation editor.
Is this the same as converting MIDI to WAV or MP3?
No. Audio to MIDI estimates note instructions from a recording. MIDI to WAV or MP3 does the reverse: it renders existing MIDI instructions into an audio waveform.

More MIDI tools