Squidbrain Audio DSP

Portable, dependency-free DSP used by Squidbrain clients. Audio is planar Float32Array[] in the nominal range [-1, 1]; selections are frame ranges with an exclusive end. It intentionally has no UI, filesystem, Web Audio, or React Native dependency.

The package is published publicly as @monosonic/audio-dsp. Release builds pin an exact package version so EAS and Cloudflare builds remain reproducible.

Noise cleanup

reduceNoiseChannels() is the general-audio cleanup path. It preserves planar channel layout and applies one linked spectral mask across stereo channels, so the stereo image does not shift as the noise floor is reduced. It is deliberately conservative and intended for steady room tone, hiss, or similar background noise in field recordings and samples.

Pass a short noiseRange containing only noise whenever possible. Without one, the function only uses an automatic profile when it can identify a genuinely quieter part of the sample; it throws rather than guessing on uniformly musical audio. Callers should provide A/B preview and make cleanup an explicit action.

Clean multimode filter

renderFilter() provides the shared clean LP/BP/HP/Focus renderer. It preserves planar channel layout and frame count, supports 12/24 dB slopes, and applies peak protection only when output would exceed the Squidbrain ceiling.

filterResponse() returns the corresponding TPT response for drawing UI. The response and audio renderer therefore use the same recipe and coefficient mapping rather than a decorative approximation.