# Clipping and attenuation examples

Three original synthesized learning assets, dedicated under CC0 1.0:
https://creativecommons.org/publicdomain/zero/1.0/

Each file is mono, 48,000 Hz, 16-bit PCM WAV, 192,000 frames (4 seconds),
384,044 bytes. No recorded performance, third-party source or AI output is used.
This dedication covers these three WAVs and their measurements, not site software
or other audio assets. These files are not part of the separate 23-WAV ZIP.

- clean.wav: 440 Hz sine, amplitude 0.25, 960-frame linear outer fades.
- clipped.wav: double the stored clean samples, hard-limit at ±0.25, write PCM16.
- quieter.wav: multiply the stored clipped samples by 0.5, write PCM16.

Reproduce: node --import tsx scripts/generate-clipping-examples.ts
The site WAV writer quantizes each step; samples below zero use 32768 and
nonnegative samples use 32767 before rounding. Decode measurements with 32768.
measurements.json records file hashes, stored sample peaks, and 220 consecutive
samples starting at second 1 for plots, all on the same vertical scale.

The ±0.25 clipping threshold intentionally leaves peaks below file full scale.
Turning down that damaged copy lowers its peak but retains its flattened shape.
These examples are not loudness matched, a real-world recording, a declipping
algorithm test, a hearing test, or a claim about recoverable speech quality.
Start at a comfortable listening volume. Do not increase volume to chase damage.