Re-run as often as you like
Five model parameters, exposed and free. Try a different threshold in seconds without spending another credit.
In-browser · private · no upload
MP3 to MIDI Converter
Drop in an MP3 and get editable notes in about ten seconds — converted privately in your browser, then cleaned up with five sliders. High-precision stem separation is an optional server feature.
Drop an audio file here
MP3, WAV, OGG, FLAC, M4A · up to 10 minutes · 3 minutes on mobile
Or try a sample:
MIDI you don't have to tidy up
BPM detection, quantization, short-note removal, velocity smoothing and hand splitting run before you download.
Every format, one conversion
.mid, MusicXML, print-ready PDF score, chord chart and ABC — no second purchase for notation.
What affects audio-to-MIDI accuracy?
There is no honest single accuracy score for every recording. Arrangement, room noise and overlapping instruments change the result, so use this guide to set expectations before converting.
AUDIO SOURCEEXPECTED RESULTWHYBEST NEXT STEP
Solo piano or guitarBest starting pointOne clear pitched instrumentAdjust onset and note length
Single voice or leadGood for a clean melodyMostly one note at a timeNarrow the frequency range
Full song mixLimited without stemsParts and harmonics overlapSeparate the target stem first
How MP3 to MIDI conversion works
Your file is decoded to 22,050 Hz mono in the browser and fed to Basic Pitch, Spotify's open-source transcription model. The model returns pitch activation and note onset data, which Tonera turns into editable notes with pitch, start, length and velocity. Because the model runs on your machine, adjusting the clean-up controls is fast and private.
Three steps
1
Drop your file in
Use MP3, WAV, OGG, FLAC or M4A up to ten minutes. Decoding and transcription happen locally.
2
Tune the clean-up controls
Kept notes are blue and discarded notes stay visible in orange, so every threshold is easy to understand.
3
Export what you need
One conversion gives you MIDI, MusicXML, a PDF score, a chord chart and ABC.
Where it works, where it doesn't
Solo piano, guitar, single-voice singing and monophonic leads transcribe well. Full mixes are harder because a pitch model cannot reliably separate bass, pads and vocals. For finished songs, use high-precision mode to separate stems first.