What we tested
We generated three 22,050 Hz mono WAV files with the same isolated C4, E4, G4, and C5 sine tones. Each note lasted 0.75 seconds, with a 0.35-second gap. The clean file peaked at 0.58. The other files used deterministic hard clipping at 0.30 or 0.10 and were scaled back to the same 0.58 peak, so the comparison changed waveform shape rather than peak level.
All three files went through the current local converter with its default controls. Each browser run returned HTTP 200, completed without console errors, and exported one musical MIDI track.
| Input | MIDI events | Detected pitches | MIDI size |
|---|---|---|---|
| Clean | 6 | C4, E4, G4, C5 | 105 bytes |
| Moderate clipping | 6 | C4, E4, G4, C5 | 105 bytes |
| Severe clipping | 5 | C4, E4, G4, G2, C5 | 99 bytes |
What changed in the MIDI
The clean result kept the four intended pitches but split C4 and E4 into a long event plus a 0.070-second edge fragment. Moderate clipping kept the same four pitch classes and the same total event count, but moved the starts or ends of several notes and raised the main-note velocities from 78–83 to 100–108.
Severe clipping removed those short C4 and E4 fragments, lengthened each main note slightly, and added a 0.360-second G2 event beneath G4. The source contained no G2. The test therefore shows that matching peak level alone did not make the transcriptions equivalent after the waveform was clipped.
What to do with a clipped source
- Use an unclipped export when you have one. Lowering the volume after clipping does not restore the flattened waveform peaks.
- Check octave-related notes. In this test, the false note appeared two octaves below the intended G4.
- Compare note ends, not just pitch names. Moderate clipping changed boundaries even though it kept the same four pitch classes and event count.
- Treat velocity as a draft. Peak-matched files produced different MIDI velocities, so source distortion can change the dynamics written into the MIDI.
What this test does not prove
It measures two exact hard-clipping thresholds on four synthetic notes. It does not set a safe loudness limit or predict results for tube saturation, guitar pedals, soft clipping, streaming normalization, vocals, real instruments, or dense songs.
For unrelated steady contamination, see the background-noise test. If the output already contains unwanted fragments, use the extra-note cleanup guide.