Free online

MP3 to MIDI Converter

Drop in an audio file and get a MIDI file back. Any instrument, chords and all, transcribed by a neural network that runs inside this page — your recording is never uploaded anywhere.

  • 12audio and video formats
  • Any instrument, polyphonic
  • Runs in your browser — nothing uploaded
How it works

Drop an audio file here

or choose a file

MP3, WAV, M4A, AAC, FLAC, AIFF, OGG — and the audio inside MP4 or MOV

Your file never leaves this browser. Nothing is uploaded.

Advertisement
  • Nothing is uploaded The model runs on your own machine. Your recording never touches a server.
  • No length limit A thirty-second riff or a nine-minute take. There is no clip cap and no trial.
  • Free, no account No signup, no credits, no watermark on the file you download.

A transcription you can argue with

The audio is analysed once. Every setting after that re-picks the notes instantly, so you can hear the difference instead of guessing at it.

Hear it, change it, hear it again

Most converters hand you a file and leave you to open a DAW to find out whether it was any good. Here the transcription is drawn as a piano roll you can play back, next to the controls that produced it. The audio is analysed once; every setting after that re-picks the notes in a few milliseconds, so you can drag the sensitivity slider and hear the difference immediately rather than uploading your song again.

  • Any instrument, not only piano

    The engine was trained without assuming what it is listening to, so a guitar riff, a bass line, a sung melody, a violin or a full piano part all transcribe. Older browser converters — including the one that used to run on this page — used a piano-only model, which is why everything came back sounding vaguely like a piano.

  • Chords, not just the top line

    Transcription is polyphonic: several notes sounding at once are written as several notes, so block chords, arpeggios and two-hand piano parts survive the trip instead of being flattened into a single melody.

  • Your audio never leaves the browser

    There is no upload step, because there is no server to upload to. The model is fetched once, then the analysis happens on your machine — which is also why it keeps working on an unreliable connection, and why nothing you convert can end up in someone's training set.

  • Pitch bends, vibrato and slides

    The model tracks pitch continuously rather than snapping straight to semitones, so a guitar bend or a singer's vibrato is written into the MIDI as bend events. Switch it off for keyboards, where it only adds noise to the file.

  • Opens at the right tempo

    The tempo is estimated from the notes and written into the file, so it lands in your DAW with the bar lines roughly where they belong. You can type over the number, and snap the timing to a grid if the source is a loop rather than a performance.

  • Twelve input formats

    MP3, WAV, M4A, AAC, FLAC, AIFF, OGG, Opus and WebM, plus the audio track inside an MP4, M4V or MOV. Decoding is handled by the browser itself, so anything it can play, this can transcribe.

From audio to MIDI in three steps

  1. Drop in your audio

    Drag a file onto the panel or pick one. MP3, WAV, M4A, FLAC, AIFF, OGG and video files all work. Nothing is sent anywhere.

  2. Say what is in it

    Piano, guitar, bass, voice — or leave it on Anything. The choice sets the note range and how strictly notes are picked out, and it can be changed after the fact.

  3. Listen, adjust, download

    Play the transcription against the original, move the sensitivity until the notes match what you hear, then download a standard .mid file.

Looking for something different?

This tool listens to a recording and writes out the notes it hears. If you already have a MIDI file, or your audio is on YouTube, start somewhere else.

How the transcription engine works

Spotify’s Basic Pitch, running on your own machine.

The model

Basic Pitch, an open-source transcription model published by Spotify under the Apache 2.0 licence. It is small and instrument-agnostic by design: rather than learning one instrument in detail, it predicts three things for every 11-millisecond frame of audio — which pitches are sounding, which are starting, and where the pitch sits between the semitones.

Model
Spotify Basic Pitch
Download
917 KB, fetched once
Analysis rate
86 frames per second
Note range
A0 to C8
Polyphony
Unlimited

Where it runs

Everything happens in this tab. The audio is decoded by the browser, downmixed to mono and resampled to 22.05 kHz, then passed through the model using WebGL on your graphics hardware. No part of the file is transmitted. Without graphics acceleration the same work falls back to plain JavaScript, which is far slower — the page will tell you if that happens. The very first conversion after opening the page is slower than the ones after it: your browser has to compile the shader programs the model runs on, which happens once. That work is started as soon as you choose a file, so most of it is over before you press convert.

What the settings change

The expensive step is inference; turning its output into notes is arithmetic. That is why the controls respond instantly.

  • Sensitivity moves the onset and frame thresholds together — how confident the model has to be before a note is written down.
  • Shortest note discards anything briefer, which is how most spurious notes are removed.
  • The instrument choice sets a frequency floor and ceiling, which is what stops a bass line being doubled an octave up by its own harmonics.
  • Pitch bends are read from the continuous pitch contour and written as bend events.
  • Snap to grid rounds start times and durations to a division of the detected tempo.

What it is not good at

Worth knowing before you judge a result. Drums and percussion have no pitch to transcribe, so they come out as noise. A dense full mix will produce a dense, approximate transcription — the model hears everything at once and does not separate instruments. Heavy distortion, heavy reverb and very quiet passages all reduce accuracy. A clean recording of one instrument gives by far the best result.

Frequently asked

Is this MP3 to MIDI converter free?

Yes, completely. There is no account, no credit system, no length cap and no watermark. The tool is supported by advertising on the page, not by charging for conversions.

Can AI really convert MP3 to MIDI?

It can transcribe, which is not quite the same thing. An MP3 is a recording of sound; MIDI is a list of notes. There is no way to reverse one into the other exactly, so a neural network listens and writes down the notes it is most confident about — the same job a musician does by ear, with the same scope for disagreement. On a clean solo instrument the result is usually close enough to edit rather than replay. On a dense mix it is a sketch.

Is there a free way to convert MP3 to MIDI online without installing anything?

This page is one. It needs no download and no plugin: the model runs inside your browser, so the only thing that gets fetched is the model file itself, once.

How accurate is the conversion?

It depends almost entirely on the recording, not on the settings. As a rough guide:

  1. Solo piano, guitar or voice, cleanly recorded — most notes correct, some timing to tidy up.
  2. Solo instrument with reverb or distortion — pitches usually right, note starts and ends less so.
  3. Full band mix — the loudest line comes through, quieter parts are approximate.
  4. Drums, percussion, spoken word — not transcribable; there is no pitch to find.
Which audio formats can I convert to MIDI?

MP3, WAV, M4A, AAC, FLAC, AIFF, AIF, OGG, OGA, Opus and WebM, plus the audio track of MP4, M4V and MOV video files. Decoding is done by your browser, so in practice anything it can play is accepted.

Is there a file size or length limit?

No fixed one. The audio is held in memory rather than uploaded, so the practical limit is your device's memory — several minutes of audio is routine on a laptop, and longer files simply take longer to analyse. There is no thirty-second trial clip and no three-minute cap.

Is my audio uploaded to a server?

No. There is no upload endpoint. The file is read by the page, decoded, and analysed on your own hardware. You can confirm it in your browser's network panel: after the model is fetched, no request carries your audio.

Can it convert drums or a beat to MIDI?

Not usefully. This model transcribes pitch, and drums are largely unpitched — it will hear the noise of a snare as a scatter of vague notes. Drum transcription is a different problem needing a different model, and no browser tool does it well today. If you need a drum MIDI part, programming it by ear from the audio is still faster than cleaning up a bad transcription.

Can it separate one instrument out of a full song?

No — that is stem separation, a separate step. This tool transcribes what it hears as a whole, so on a full mix it tends to follow whatever is loudest and clearest. If you want just the bass line from a finished track, run a stem separator first and convert the isolated stem here.

Why does my transcription have notes that are not in the recording?

Usually harmonics or reverb tails being read as real notes. Two fixes, both instant: lower the sensitivity so only clearly struck notes survive, and raise the shortest-note setting so brief artefacts are discarded. Choosing the right instrument helps too, since it puts a ceiling on the frequencies considered — this is the single most effective fix for a bass line that has been doubled an octave up.

Can I convert a YouTube video to MIDI?

Use the YouTube to MIDI converter, which takes a link directly. Otherwise you would need to obtain the audio yourself first and drop it here.

Can I get sheet music instead of a MIDI file?

Convert here first, then run the .mid through the MIDI to sheet music converter for notation you can read or print. Quantising to a grid before you download usually produces far more legible notation.

How do I convert MIDI back to MP3?

The other direction is the MIDI to MP3 converter, which renders a MIDI file to audio with an instrument of your choosing.

Does it work on a phone?

It runs, but it is slow: phone browsers give a page far less graphics performance than a laptop does, and the analysis is the heavy part. For anything longer than a short clip, a desktop browser is worth the wait it saves.

What is the best software for converting MP3 to MIDI?

For a single clean instrument, a browser tool like this one is usually enough and costs nothing. For production work on dense material, paid desktop software with stem separation and manual editing built in — Melodyne, or the audio-to-MIDI in a full DAW — gives you more control over the mistakes, which is really what you are paying for. Everything available today gets the same class of result on the same class of recording; the difference is how much editing the tool lets you do afterwards.