What we tested
We generated a 22,050 Hz, 16-bit mono WAV containing four isolated sine tones: C2, C3, C4, and C5. Each tone lasted 0.900 seconds, with 0.300-second gaps and the same 0.68 peak amplitude. We kept one file unfiltered and processed two copies with second-order Butterworth-style high-pass filters at 180 Hz and 400 Hz.
We ran the three files through the current local converter at its default controls, downloaded each MIDI file, and parsed its note events. The converter logic was not changed.
| Source condition | MIDI events | Intended pitches retained | MIDI size |
|---|---|---|---|
| Unfiltered | 14 | C2, C3, C4, C5 | 170 bytes |
| 180 Hz high-pass | 8 | C2, C3, C4, C5 | 121 bytes |
| 400 Hz high-pass | 5 | C2, C3, C4, C5 | 97 bytes |
The cutoff did not act like a MIDI note gate
A high-pass filter attenuates frequencies below its cutoff gradually; it does not delete every lower pitch. That distinction appeared in the exported files. Even the 400 Hz version retained C2, C3, and C4, although all three fundamentals sit below 400 Hz.
The clearest measured change was event structure. The unfiltered C2 was divided into seven adjacent fragments. The 180 Hz version reduced C2 to two fragments, and the 400 Hz version reduced it to one. C3 changed from four fragments to three and then two. C4 changed from two events to two and then one, while C5 stayed one event in every file.
Why fewer events is not automatically a better transcription
The filtered outputs were shorter and less fragmented for this synthetic source, but that does not establish a cleanup recipe. Filtering changes the audio supplied to the pitch model. On a real instrument it can weaken fundamentals, emphasize overtones, change attacks, or remove wanted bass content.
Judge the result against the source, not only by its event count. A five-event MIDI can still have a wrong boundary or unwanted split, while a larger draft may preserve a performance detail you need.
How to test EQ before conversion
- Keep the original. Export a filtered copy instead of overwriting the recording.
- Change one control. Compare the same passage with and without one high-pass setting.
- Inspect low notes and octaves. Make sure a cleaner-looking result did not lose bass pitches or substitute overtones.
- Compare timing and velocity too. Matching pitch classes do not mean the MIDI files are identical.
What this test does not prove
It does not define an ideal cutoff, measure a real mix, or predict other filter slopes, instruments, compressors, video encodes, or converters. It records the current tool's output for three controlled files and shows that source EQ can alter MIDI event boundaries without simply erasing every pitch below the cutoff.