What we tested
We generated three 22,050 Hz mono WAV files containing the same isolated C4, E4, G4, and C5 sine tones. Each dry note lasted 0.65 seconds, and note starts were 1.10 seconds apart. One file stayed dry. The other files added five decaying repeats spaced 120 ms or 320 ms apart, with each repeat at 55% of the previous level.
All three files went through the current local converter with its default controls. Every browser run returned HTTP 200, completed without console errors, and exported one musical MIDI track.
| Input | MIDI events | Pitch classes | MIDI size |
|---|---|---|---|
| Dry | 7 | C4, E4, G4, C5 | 113 bytes |
| 120 ms echo | 29 | C4, E4, G4, C5 | 290 bytes |
| 320 ms echo | 24 | C4, E4, G4, C5 | 258 bytes |
What changed when echoes were added
The dry result preserved the four intended pitch classes but added three short edge fragments. The 120 ms condition turned C4, E4, G4, and C5 into 8, 6, 8, and 7 events. The 320 ms condition produced 7, 6, 6, and 5 events for those pitches.
The longer delay also kept earlier pitches sounding into later source notes. For example, C4 events overlapped the E4 section, and an E4 event overlapped the G4 section. No unexpected pitch class appeared in this exact test; the measured problems were fragmentation and overlapping tails.
How to handle an echo-heavy source
- Start with the driest copy you control. A direct recording gives the detector clearer note boundaries than a speaker recorded in a room.
- Reduce delay or room tail before transcription. Listen after processing so cleanup does not remove quiet intended notes.
- Inspect repeated same-pitch events. Merge them only when the source contains one note followed by effect repeats, not deliberate retriggers.
- Check overlaps at each note change. An older echo can remain under the next pitch and create simultaneous MIDI notes.
What this test proves—and what it does not
It records how the current converter handled two exact multi-tap echo patterns on four synthetic notes. It does not establish a safe delay time, a dereverberation score, or expected accuracy for live rooms, plates, springs, stereo delays, vocals, instruments, or complete songs.
For steady noise rather than repeated tails, see the background-noise test. If the result already contains short fragments, use the extra-note cleanup guide.