Published by PitchChanger · Updated · How these guides are made
How to use this workflow
Inspect an unchanged download
Keep the downloaded WAV and inspect a copy in an editor. Measure sample peaks, identify the location of the suspected distortion and compare that passage with the source mix. Keep monitoring quiet and consistent. Do not use speaker loudness as a peak meter.
Why attenuation cannot restore a clipped shape
Consider hypothetical normalized samples 0.8, 1.2, 0.8. A hard clip at 1 turns them into 0.8, 1.0, 0.8. Reducing the clipped result by half gives 0.4, 0.5, 0.4; reducing the unclipped sequence gives 0.4, 0.6, 0.4. These calculations demonstrate lost shape, not measurements of the separator.
Swipe or scroll sideways to read all columns.
| Stage | Sample 1 | Sample 2 | Sample 3 |
|---|---|---|---|
| Unclipped values | 0.8 | 1.2 | 0.8 |
| Hard clipped | 0.8 | 1.0 | 0.8 |
| Clipped then halved | 0.4 | 0.5 | 0.4 |
Choose the appropriate response
If overload occurs only when tracks are summed, lower levels before the overloaded stage and export again. If the same distorted character is already in the source, turning down a separated stem will not reconstruct the missing waveform. If it appears only in an estimate, compare another suitable source or separation method; label the limitation instead of promising repair.
Keep evidence specific
Record file revision, timestamp, peak units and where in the chain the measurement occurred. Sample peak and true peak are different measurements; name the one your inspector provides. PitchChanger automatically applies output gain to its named stem exports, but that is not proof that a source or estimate contains no distortion.
Common questions
Does a peak below full scale prove the stem is clean?
No. A distorted source or processed waveform can be below full scale. Compare the suspect passage with the original and inspect the relevant stage.