How to record audio that converts well to MIDI
Automatic transcription is only as good as the recording you give it. These habits make the biggest difference, and none of them need special gear.
Record one instrument or one voice
The note model is built for a single instrument at a time. A solo piano, a guitar, a hummed or sung melody all work well. A full band mix makes the model guess which notes belong to which part, so expect extra or missing notes there.
Keep the room quiet and the level steady
Background noise, fans and room echo turn into short, quiet ghost notes. Record close to the source, in a small furnished room if you can, and keep the input level well below clipping.
- Phone: hold it 20–40 cm from the instrument or your mouth.
- Audio interface: aim for peaks around −12 dB.
- Turn off effects such as reverb, delay and auto-tune before recording.
Play or sing clearly
Separate repeated notes slightly, so each one has a clear start. For humming, use a soft consonant like “da” or “du” on every note; a continuous “mmm” blurs where notes begin.
Short takes beat long ones
Record the part you need, not the whole song. Shorter files convert faster (and the tool accepts up to 5 minutes on a computer, 2 minutes on a phone). MP3 or WAV both work.
Then check it against the original
Convert, play the notes together with your recording, and fix what’s off before exporting. The guide on fixing and exporting shows how.