If you recorded it yourself, upload the WAV rather than an MP3 export. Uncompressed audio gives the transcriber a cleaner spectrum, and you can hear the difference in the note count.
MP3 encoding throws away what a listener is unlikely to notice — and some of what it throws away is exactly what a transcription model looks at. Lossy compression smears note onsets, and above roughly 16 kHz it removes content entirely, which blurs the attack transients that tell the model where one note ends and the next begins. Quiet notes under a loud chord are the first casualties.
A WAV keeps all of it. Feed the transcriber the original take from your DAW, field recorder or phone rather than a bounced MP3, and onsets land more precisely. The engine is the same one behind our MP3 to MIDI converter — tuned onset and frame thresholds, plus a harmonic-ghost filter that removes the phantom octave notes raw models invent under sustained chords.
The model hears what your microphone heard. A few things help more than any setting on our end:
Want a readable score rather than a MIDI file? Audio to sheet music renders the same transcription as notation.