Basic Pitch MP3 to MIDI Converter
Run Spotify's Basic Pitch neural network on your MP3 or WAV — the same open-source model, with a piano-roll preview and fine-tune controls, right in your browser.
- Spotify Basic Pitch
- Piano roll preview
- Fine-tune controls
Click a note to hear it
How it works
Upload your audio
Drop an MP3 or WAV. The Basic Pitch model transcribes the pitches and timing into MIDI — no account needed to convert and preview.
Preview and fine-tune
Inspect the transcription on a piano roll and adjust the same parameters Basic Pitch exposes — onset/frame sensitivity, minimum note length, and pitch bends — then re-transcribe.
Download MIDI or sheet music
Export the MIDI, edit notes on the roll, or export sheet music as MusicXML/PDF for MuseScore, Sibelius, or printing.
Formats & limits
- Input formats
- MP3
- Output
- MIDI
- File limits
- 50 MB · 10 min
- Price
- Free · no sign-up
Getting the most out of Basic Pitch
- Basic Pitch is strongest on monophonic, harmonic instruments — vocals, flute, violin, lead guitar, piano melodies.
- Lower the sensitivity thresholds if you get too many spurious notes; raise them to catch quiet or short notes.
- Enable pitch bends for expressive sources like voice or bent guitar strings so slides are captured.
- The model outputs a draft — use the editor to fix the last few notes rather than re-recording.
Frequently asked questions
What is Basic Pitch?
Basic Pitch is an open-source audio-to-MIDI transcription model that Spotify released alongside a 2022 ICASSP research paper, built specifically to be instrument-agnostic rather than tuned for one instrument type. It listens to raw audio and detects three things at once — which pitches are sounding, when each note starts and ends, and continuous pitch-bend movement like slides or vibrato — without needing separate training for vocals, guitar, piano, or any other source. That instrument-agnostic design is also its main trade-off: research comparing it against models built for one specific instrument shows it usually loses on raw accuracy to those specialists, in exchange for working reasonably across everything without retraining. See our transcription accuracy benchmarks for the actual published numbers by instrument, including where it wins and where it loses to specialized tools.
Is this the real Basic Pitch model?
Yes — transcription here runs the actual official Basic Pitch model from Spotify's ICASSP 2022 release, not a re-implementation, approximation, or a different model marketed under the same name. What this site adds on top is the surrounding product: a browser-based upload flow, an interactive piano roll to preview and compare the result against your original audio, fine-tuning controls that map directly to Basic Pitch's own exposed parameters, note-level editing, and export to MIDI or MusicXML/PDF sheet music. The transcription quality and its known strengths and limitations are therefore identical to what Spotify's own published research describes — nothing about running it through this interface changes the underlying model's behavior, its strengths, or its known limitations.
Is it free to use?
Yes — converting and previewing is completely free, with no account required at any point. Upload an MP3 or WAV, watch Basic Pitch transcribe it, and listen back against the original immediately. What requires signing in is everything downstream of that first pass: saving results to a persistent conversion history, using the full note-editing toolkit rather than just previewing, and exporting the corrected result as MusicXML or PDF sheet music alongside the plain MIDI download. In practice this means you can fully evaluate whether a transcription is usable before ever deciding whether creating an account is worth it, and a quick one-off MIDI download needs no account at all — nothing about the free path runs a reduced version of the model.
What parameters can I adjust?
You can tune the same three parameters Basic Pitch itself exposes, rather than a simplified subset: the onset and frame detection thresholds that control how sensitive the model is to picking up a new note versus staying silent, the minimum note length used to filter out very short spurious detections, and whether to enable multiple simultaneous pitch bends for expressive sources like voice or bent guitar strings. Adjusting any of these and re-transcribing reuses the same uploaded audio, so you can try several settings in sequence — tightening thresholds to remove ghost notes, or loosening them to catch quiet passages — without uploading the file again each time, which makes trial and error on a difficult recording much less tedious.