What we tested
We generated three 22,050 Hz, 16-bit stereo WAV files containing the same isolated C4, E4, G4, and C5 sine-wave sequence. Each note lasted 0.900 seconds and each gap lasted 0.300 seconds. The left channel stayed fixed; the right channel used the same signal with a delay of 0, 5, or 20 milliseconds.
Pitch, source peak, note timing, envelopes, sample rate, converter controls, and converter code stayed fixed. Only the right-channel delay changed.
| Right-channel delay | Downmixed RMS | MIDI events | Result |
|---|---|---|---|
| 0 ms | 0.33198 | 6 | Four intended pitches; C4 and E4 each had a short tail |
| 5 ms | 0.21225 | 6 | Four intended pitches; C4 and G4 each had a short tail |
| 20 ms | 0.19808 | 7 | Four intended pitches; short tails after C4, E4, and G4 |
All intended pitches survived
Every output retained C4, E4, G4, and C5. The 0 ms and 5 ms files each produced six events in a 105-byte MIDI file. The 20 ms file produced seven events in 113 bytes.
The delay changed where notes were split
With aligned channels, the converter added short tails after C4 and E4. At 5 ms, the extra tails followed C4 and G4. At 20 ms, short tails followed C4, E4, and G4. Main-note durations also moved by up to 0.046 seconds between these three outputs.
Velocity changed even though channel peaks stayed fixed
The main C5 velocity fell from 79 with aligned channels to 65 at 5 ms and 56 at 20 ms. The right-channel delay changed the waveform created when the stereo channels were averaged: measured mono RMS fell from 0.33198 to 0.21225 and 0.19808.
A time offset creates frequency-dependent reinforcement and cancellation during downmix. These measurements show correlation between the changed downmix and changed MIDI output; they do not identify one universal delay threshold.
What to do with widened stereo audio
- Compare a mono copy. Keep the original, then test a careful mono render before transcription.
- Inspect note boundaries. Merge short same-pitch tails when they belong to one sustained performance.
- Check velocities by ear. MIDI velocity can change when stereo summing changes the analyzed level.
- Use an isolated stem when possible. A clean source makes it easier to judge whether a MIDI fragment is musical or accidental.
Limits of this result
These are synthetic sine tones with identical left and right content except for delay. The test does not cover different stereo performances, panning automation, room microphones, chorus, or every downmix method. It records three exact outputs from the current local converter at default controls.