# Original high-pass filter listening example

Created by Saturalabs on October 24, 2026. These are original synthesized learning
assets, not recorded wind, speech, music or AI separation results.

Both WAVs contain 192,000 mono frames: four seconds at 48 kHz, 16-bit integer PCM,
384,044 bytes including the basic header. `original.wav` combines a 25 Hz cosine
at amplitude 0.18 and a 440 Hz cosine at amplitude 0.12. A linear 20 ms fade is
applied at each outer edge. Each generated sample is clipped/rounded to signed
PCM16 using the project's `pcm16Value`, then written directly into the WAV.

`80hz.wav` is the actual output of `lib/audio/high-pass.ts` from those stored
PCM values, decoded as integer / 32768. It uses the same processor as the public
high-pass tool: cutoff 80 Hz, Q=1/sqrt(2), a second-order Butterworth response,
12 dB/octave asymptotic slope, double-precision state, one channel, zero initial
state, and one final PCM16 conversion. No normalization, extra fade or appended
filter tail is applied. Both files have the same stored duration; phase and
outer-edge behavior change. No prepared sample exceeded full scale.

## Measured tone amplitudes

`measurements.json` projects the actual stored PCM samples onto sine and cosine
at each known frequency over source seconds 1–3, away from outer fades and
startup. Magnitude is 2 * sqrt(cosine_sum^2 + sine_sum^2) / sample_count.
Change in dB is 20 * log10(filtered_amplitude / original_amplitude).

- 25 Hz: approximately −20.25 dB.
- 440 Hz: approximately −0.005 dB.

Public report: https://www.saturalabs.com/audio-examples/high-pass/measurements.json

The report also includes the separately calculated steady-state Butterworth
magnitude for comparison and the SHA-256 of each WAV. The low tone becomes
smaller; it is not completely removed. These component-amplitude readings are
not integrated loudness, true peak, SNR, percentages of wind removed or a speech
quality score. Playback can resample or change gain, and some speakers cannot
reproduce the lowest tone usefully. A pure-tone filter example does not establish
that a damaged word can be repaired.

Regenerate in the project with:

    node --import tsx scripts/generate-high-pass-examples.ts

Native verification independently decodes these exact files with FFmpeg and
compares the filtered samples with its two-pole high-pass implementation using
the same cutoff and Q. The reproducible signals describe this particular filter,
not every editor's high-pass implementation or arbitrary real-world recordings.

## Reuse

To the extent possible under law, Saturalabs dedicates these two original WAVs
and their measurement report to the public domain under CC0 1.0:
https://creativecommons.org/publicdomain/zero/1.0/

No attribution is required. This dedication covers these learning assets, not
third-party recordings, website software, fonts or AI outputs generally.