Why Full-Song Audio to MIDI Fails — and How to Fix It
Last updated July 15, 2026
The most common way people try audio to MIDI is also the one that fails: they drop in a finished song and get back a cloud of wrong notes. It is not a bug in the tool — it is the hardest unsolved problem in the field. The good news is the fix is simple and it is what the pros already do: convert one isolated instrument at a time. Here is why a full mix breaks conversion, and how to get a clean result instead.
Drop in one isolated instrument — a bassline, a topline, a guitar — and get a clean, editable .mid. Free in your browser.
Try the converter free →The #1 mistake: feeding it a full mix
A full song is dozens of notes sounding at once — kick, snare, bass, chords, a vocal, effects — all summed into a single waveform. When you ask a converter to transcribe that, it has to find every pitch, decide where each note starts and stops, and figure out which note belongs to which instrument, all at the same time. The result is the tangle everyone recognizes: phantom notes, the bass an octave off, the melody buried, timing all over the place.
Why polyphony is so hard
Monophonic audio — one note at a time — is the tractable case: the model only has to track pitch and timing. Polyphonic audio is a different problem entirely. Simultaneous notes overlap in both time and frequency, and their harmonics interfere, so a partial of one note can masquerade as the fundamental of another. Untangling that from an arbitrary mix is still considered an unsolved research problem.
The numbers show the cliff. In the 2025 AMT Challenge, the best system scored a note-level F-measure around 0.72 on a single instrument — and around 0.44 once three instruments played together. This is exactly why Spotify's Basic Pitch, the open engine many tools use, states plainly that it works best on one instrument at a time.
The fix: isolate one instrument first
Give the converter a soloed part and the hard problem disappears — it only has one line to follow. If you only have the full song, split it into stems first (a stem separator or your DAW's splitter), export the single part you want, and convert that. Practitioners are unanimous that a piano-only stem transcribes far better than a full mix, and a clean studio recording beats a phone capture.
Then convert the isolated part in the audio-to-MIDI converter — that is the whole reason it asks for one instrument rather than a song. For the step-by-step, see how to convert MP3 to MIDI.
Part by part: what to watch
- Bass is notorious for landing an octave off and picking up phantom notes — convert it soloed and check the octave.
- Drums are a separate problem: a drum transcriber classifies hits (kick, snare, hi-hat), it does not read pitches, so it needs its own path.
- Chords and dense synths are the hardest — expect the most cleanup, or play them in by hand.
- Vocals, leads, and single-note lines are the easiest and convert cleanest — start here.
How Selekt is built for this
Selekt's converter asks for one isolated instrument on purpose — it is the honest way to get a usable result. Guitar, vocal, and melody parts convert free and on-device in your browser; bass, piano, and drums use heavier cloud models. And if you genuinely have a whole song to break down, the stem separator splits it and extracts MIDI for every part in one pass, so you get the stems and their MIDI together instead of fighting a single mix.
Isolate the instrument you want, convert it clean, then re-voice it on cleared sounds and build.
Open the audio-to-MIDI converter →Frequently asked questions
Can you convert a full song to MIDI?
Not cleanly. A full mix is polyphonic — many notes at once — which is the hardest case for audio-to-MIDI and produces tangled, wrong-note results. Split the song into stems and convert one isolated instrument at a time instead.
Why is my converted MIDI full of wrong notes?
Almost always because the source was a mix or a dense, chordal part. Overlapping notes confuse the pitch estimator, so it hallucinates notes and misplaces octaves. Feed it a single isolated, monophonic line and the result gets dramatically cleaner.
How do I convert just one instrument from a song to MIDI?
Run the song through a stem separator to isolate the part (the bass, the vocal, the guitar), export that single stem, then convert the stem to MIDI. On Selekt the stem separator can extract MIDI for every stem in the same pass.
Do I have to split stems before converting?
For anything more than a soloed instrument, yes. If your source is already one clean part, you can convert it directly. If it is a mix, splitting stems first is the difference between a usable MIDI file and junk.
Why is my bass the wrong octave after conversion?
Low frequencies are ambiguous for pitch estimators — the fundamental and its harmonics are close together — so bass often lands an octave high or low. Convert the bass soloed and expect to nudge the octave; it is a quick fix in any DAW.
Key takeaways
- A full mix is polyphonic — many notes at once — which is the hardest, still-unsolved case for audio to MIDI.
- Accuracy collapses as parts overlap: about 0.72 F-measure on one instrument, about 0.44 on three (2025 AMT Challenge).
- The fix is to isolate one instrument first — split stems, then convert the single part.
- Bass drifts octaves, drums need their own path, chords are hardest, single lines convert cleanest.
- Selekt's converter asks for one isolated part on purpose; its stem separator can split a whole song and extract MIDI in one pass.
