Do drums make MIDI conversion worse?
Not before chords do in this controlled source check. The melody-only fixture retained 7 of 8 notes; adding chord triads reduced retention to zero, and adding drums afterward did not change the median.
Controlled mixture check
| 8-note target source | Median across 10 fixtures | What changed |
|---|---|---|
| Melody only | 7 retained; 1 missed; 0 extra | Baseline single-line result |
| Melody + chord triads | 0 retained; 8 missed; 3 extra | Chord competition entered |
| Melody + chord triads + kick and high percussion | 0 retained; 8 missed; 3 extra | No further median change |
What to change in the source
This does not say drums never matter. It says they were not the variable that changed this fixture after chords were already present. Choose an exposed vocal, solo, or single-note phrase, or isolate the target line before conversion.
Fixture conditions and limits
Each level used 10 generated mono 44.1 kHz 16-bit PCM WAV fixtures. The target was an eight-note melody; chord levels added three simultaneous tones per target note; the third level added a 90 Hz kick and high-percussion pulse. Balanced Melody / solo instrument mode used a 2048-sample analysis window and a 1024-sample hop. A retained note needed one output within 50 cents and at least 50% overlap. Detector SHA-256: DDA185AC8FB6A179A4FCFA73FD4197ACC95B8B6F59B233F5D3E330F596309F9F.
These are controlled source checks, not a complete-song, stem-separation, chord-transcription, drum-transcription, or universal accuracy promise.