Clean melodies are easier than dense mixes
A solo piano line, guitar melody, bass phrase, humming voice, or isolated instrument generally gives the transcription model clearer evidence than a mastered song containing vocals, drums, bass, pads, reverb, and several instruments at once. Polyphonic piano can work well because the pitches are stable, but complex mixes may create extra notes from harmonics or miss quiet notes hidden under louder sounds. For the cleanest editable result, start with the least crowded source you have. If you own stems, an isolated instrument stem is usually a better input than the full mix.