What converting a song to MIDI actually means
MIDI is not audio. A MIDI file is a list of instructions: this note, this loud, starting here, ending there. An MP3 or WAV is the opposite, a recording of the sound itself. So converting a song to MIDI is really transcription. Something has to listen to the audio and write down which notes it hears and when.
Not long ago that meant software that struggled with anything harder than a whistled tune. Today small AI models do it well, and some are free and open source. The one we use is Basic Pitch from Spotify's Audio Intelligence Lab. It runs in a web browser, handles voices and instruments, and can hear more than one note at a time.
What no model can do is pull a clean piano part out of a dense mix with drums, bass, synths and vocals stacked on top of each other. That one thing decides whether your result is useful, so the method below starts there.
Step 1: Get one part on its own
Note detection works best on one clear part: a vocal, a bass line, a lead synth, a piano. Feed it a full mix and it will catch the loudest notes, miss the rest, and turn some drum hits into notes nobody played.
If you made the song, this step is easy. Solo the track in your DAW and bounce it to WAV. If you only have the finished song, use a stem splitter. Our stem splitter runs in your browser and separates a song into vocals, drums, bass and the rest, so you can take the vocal or the bass stem and convert just that.
Stems from a splitter are not perfect. A little bleed from other instruments is normal, and the detector mostly ignores quiet bleed. If stray notes still show up, the sensitivity setting in step 3 handles most of them.
Step 2: Run it through a note detector
Open the Audio to MIDI converter and drop in the file. MP3, WAV, M4A and FLAC all work, and so does a voice memo straight off your phone. The file is never uploaded. The model downloads to your browser once and the analysis runs on your own computer, so there is no queue and no account. Files can be up to 10 minutes long.
Pick the source type before you judge the result. Melody or vocal keeps one note at a time, which is right for singing, humming, bass lines and lead lines. Instrument or chords lets notes overlap, which is right for piano, guitar and pads. Chord mode on a vocal gives you ghost notes stacked under the melody, and melody mode on a piano throws the harmony away.
A three minute file usually takes under a minute. When it finishes you get a piano roll of the detected notes, and you can play the MIDI back and then the original to hear where they disagree.
Step 3: Clean it up before you export
Two settings fix most bad results. Sensitivity decides how sure the model has to be before it writes a note. Turn it down if breaths, reverb tails or bleed show up as notes. Turn it up if quiet notes are missing. Shortest note throws away anything shorter than the time you set, which is the fastest way to remove the tiny blips that vibrato and slides create in a vocal.
Change one setting at a time and press Play MIDI after each change. The notes update right away, so you hear the result without running the file again. When it sounds close, press Download MIDI.
Then expect a little editing in your DAW. Even a good transcription can put a note an octave off, split a held note in two, or miss a fast run. Fixing five notes by hand is still far quicker than writing out a whole part from scratch.
Step 4: Get the MIDI into your DAW at the right tempo
The MIDI file keeps the exact timing of the audio, but it is saved at 120 BPM, because the detector does not know your song's tempo. If your project runs at 120, drag the file in and it lines up. If not, you have two options.
The simple one: find the song's tempo with the Key & BPM Finder, set your project to 120 while you import, then change the project tempo to the real one and let your DAW move the MIDI with it. Most DAWs ask whether imported MIDI should follow the project tempo, so say yes. The other option is to import at 120 and quantize afterwards, which also tightens timing that drifts a little in a live performance.
Knowing the key helps too. If a stray note sits outside the key the finder gives you, it is probably a detection mistake, and you can delete or move it with confidence.
What a song turned into MIDI is good for
The most common reason is learning a part. Turn a riff or a vocal melody into MIDI, slow it down, and see exactly which notes are played. The second is swapping the sound: take a hummed idea or a guitar line and put it on a synth or a string patch. The third is remixing, where you rebuild a melody from a sample with your own instruments instead of reusing the original recording.
If the melody only exists in your head, hum it into your phone and convert the memo. If you have nothing at all yet, the Melody Generator writes MIDI ideas from scratch.