How to convert audio to MIDI

What the converter is estimating

The audio contains a waveform, not a list of notes. The converter measures pitch over time and decides where one note ends and the next begins. It also estimates velocity from the signal. Every one of those steps is an inference, which is why the output needs review.

A clean monophonic source gives the detector a single pitch to follow. A full mix gives it several overlapping pitches, drums, reverb, and vocals at once. The result may still contain useful fragments, but it is not a reliable transcription of the arrangement.

Prepare a recording that can work

Use one voice, one instrument, or one clear melody. Reduce backing music, room echo, delay, and distortion. Trim the file so the useful phrase starts at the beginning; EditMIDI analyzes the first 30 seconds of the local draft workflow.

A dry close microphone is better than a loud processed take. If the source is stereo and the melody is panned to one side, check whether summing to mono changes the balance. A part that sounds thin in mono may confuse the pitch tracker.

What the draft contains

The draft contains estimated note pitches, start times, lengths, and velocity values for one dominant pitch at a time. It does not preserve the singer, instrument tone, effects, lyrics, stereo image, or the exact sound of the recording.

Chords, drums, dense accompaniment, and several instruments playing together are outside the monophonic detector. A piano chord may produce one note or a confusing cluster; a drum hit may produce a short low note. Treat those as artifacts to remove, not as musical information.

Clean the common errors

Octave errors happen when the detector locks onto a harmonic instead of the fundamental. The note appears twelve semitones away from where you expected. Select it and transpose it back by 12 rather than re-recording the whole phrase.

Short false notes often appear around consonants, breaths, and transient attacks. Delete them. Wide vibrato or a slide can split one held syllable into several notes; merge or resize the fragments when the phrase should be sustained. Then quantize only if the timing needs it.

Keep the original recording as the reference

Loop the draft against the source and compare phrase by phrase. The goal is not a perfect copy of the recording; it is a set of notes that captures the musical idea well enough to edit and re-instrument. Some expressive details will be lost, and that is normal.

If the draft is unusable, improve the source before trying again. A cleaner take with less accompaniment usually helps more than another pass through the same converter. For a full mix, a different workflow such as stem separation and per-instrument transcription is required.

Export and use the MIDI

When the phrase reads correctly, export a standard .mid file. Assign your own instrument in a DAW, check the tempo, and compare the result with the original. The MIDI is a starting point for arrangement, not a claim that the recording has been perfectly transcribed.

Use audio you own or have permission to transform. Local processing keeps the recording on your device, but it does not change copyright, licensing, performer, or publishing rights. The converter changes where the work happens, not who owns the material.

When stems are a better starting point

If the recording contains a full band and you need more than the melody, stem separation is a better first step than a monophonic converter. Separate the vocal, bass, or instrument you care about, then analyze that cleaner stem. The extra step usually produces fewer false notes.

Stem separation is still an estimate, and it can introduce artifacts. Use it when the source is dense and the melody is buried. For a hummed idea or a single instrument, the direct audio-to-MIDI path is simpler and keeps the workflow in one browser tab.

A worked example

Imagine a phone recording of someone humming a chorus over a quiet guitar. The voice is close to the microphone, the guitar is soft, and the phrase repeats twice. That is a good candidate: one dominant pitch, a clear start, and little percussion. The draft will still need cleanup, but it has a chance.

Now imagine a full mix with drums, bass, a wide synth, and a vocal buried under reverb. The detector has to choose among overlapping pitches and transient noise. It may produce fragments of the vocal, but the result is not a reliable transcription. Stem separation or a cleaner source is the better first step.

The difference is not the file format. It is how much competing information the recording contains. Prepare the source for the question you are asking, and the conversion has a much better chance of being useful.

• Good input: one clear pitch, dry, close, and trimmed.
• Hard input: a full mix with drums, reverb, and overlapping parts.
• Better next step for a dense mix: separate a stem first.