LocalMIDI

Audio to MIDI / Source preparation

Cleaner input gives audio to MIDI a fairer chance.

The short answer: start with one clear melodic line, trim silence and distracting noise, and avoid asking one conversion pass to separate several competing instruments. Source preparation cannot guarantee a perfect MIDI file, but it removes avoidable ambiguity before detection begins.

What “clean” means here

Clean does not mean loud at any cost. It means the intended pitch is easier to distinguish from room noise, percussion, backing harmony, and competing voices. A clipped recording can be worse than a quieter one because distortion creates extra frequency content that resembles additional notes.

For a first pass, choose a short phrase with one dominant melody. A solo vocal, whistle, monophonic instrument, or exposed lead line gives a pitch detector fewer simultaneous explanations. Chords and full mixes can still be useful experiments, but the output should be judged as a transcription of the most detectable line, not as a promise to recover every arrangement part.

A focused source leaves fewer competing signalsThree input layers narrow to one focused melodic line before conversion, then produce editable MIDI note events.mixed sourcefocused lineMIDI notesConcept diagram; not a measured accuracy result.
Original preparation schematic: reduce competing signals before conversion.

Controls that matter after preparation

LocalMIDI's detector evaluates frame amplitude, pitch confidence, and whether a positive pitch was found. Its sensitivity control is relative to the loudest analysis frame, so adding a very loud unrelated sound can make a quieter intended phrase harder to pass the gate. Normalize carefully; do not assume the biggest waveform is the useful melody.

The shortest-note control removes detected segments below its selected duration. It cannot restore a pitch that never cleared the confidence or amplitude checks. Set the expected lowest and highest notes around the source, especially after transposition, so valid notes are not discarded by an overly narrow range.

These are implementation consequences, not an accuracy benchmark. They explain why changing the source and changing a filter solve different problems.

A practical preparation pass

  1. Trim the clip to the phrase you want to inspect. Silence at the beginning and end is fine, but remove long unrelated sections so you can judge one result.
  2. Reduce competing layers where possible. For a mix, try a stem or an exposed melody before increasing sensitivity.
  3. Listen for distortion, room rumble, and consonant-heavy noise. If the source is vocal, a clear sustained vowel is easier to assess than a noisy consonant.
  4. Convert the same short phrase after one change at a time. Compare the piano roll and listen to Preview notes; keep the source that preserves the intended phrase with fewer repairs.
  5. Use the missing-note guide when events vanish, and the timing guide when events are present but misplaced. Do not use BPM controls to repair a pitch-detection problem.

Preparation is a way to make the question simpler: “which notes are in this line?” Once the input answers that question clearly, MIDI cleanup becomes more predictable.

Start with the melody you can name

Open the audio to MIDI converter with a focused phrase. For vocals, the voice guide explains why the result is note data rather than words. If short events disappear, read the missing-note guide.