Fades and gain do different jobs
A quieter edge is not the same as a loudness-normalised clip.
What a linear fade changes
A fade-in ramps the selected beginning from zero toward full gain. A fade-out takes the end toward zero. Values are entered in milliseconds and rounded to decoded sample counts. Each fade is 0–2,000 ms; their combined duration cannot exceed the selection.
Try the generated example: select 1–4 seconds and apply 200 ms in, 300 ms out, gain 0.8×. The output remains a three-second region. Fades do not add silence before or after it, move the cut or change speed.
Gain and hard clipping
Gain multiplies every selected sample after applying the edge envelope. If a value exceeds the PCM range −1 to +1, WavePatch counts it and clamps it. A count of clipped sample values can be larger than the number of frames when both stereo channels clip.
Reduce gain to avoid newly clipped values. A peak below one does not establish a particular perceived loudness or prove that the source was never distorted. There is no LUFS analysis, compressor, limiter, noise removal, normalisation or hearing-level measurement.
Compare the same region
Original selection plays the decoded region without these gain/fade settings. Processed selection plays the saved processed buffer. Apply changes before comparing. Stop playback before changing device volume; start at a comfortable low level. The app’s audio-engine status confirms software playback, not that your speakers were audible.