Skip to main content

​
songbird-dsp

Shared DSP primitives used across the engine, plugins, and state layers. This is the canonical location for all reusable audio math — plugin crates should import from here rather than implementing DSP inline.

​
Organization

The crate uses atomic design to organize DSP code:
ModulePurpose
atoms/Single-purpose DSP primitives grouped by domain
molecules/Composed building blocks (multi-atom processing chains)
backend/DspBackend trait + platform implementations (RustNativeBackend)
graph/DspNode trait + runtime graph evaluator
utildB/gain conversions, metering helpers, ballistic smoothing

​
Atoms (by domain)

  • atoms::filter — Biquad (TDF-II, Audio EQ Cookbook), SVF (Zavalishin TPT).
  • atoms::osc — Oscillators (planned: sine, polyblep, wavetable, noise).
  • atoms::envelope — ADSR envelopes (linear + exponential curves).
  • atoms::delay — Delay lines with fractional interpolation (none, linear, cubic Hermite).
  • atoms::dynamics — Dynamics processors (planned: compressor, limiter, gate).
  • atoms::modulation — Modulation sources (planned: LFO, sample & hold, slew).
  • atoms::spectral — FFT wrapper (rustfft), STFT/iSTFT, windowing, spectrum analyzer, oversampling.
  • atoms::resample — Linear + windowed sinc resampling (Blackman-Harris, up to 64 taps).
  • atoms::midi — MPE tracker and voice allocator.
  • atoms::output — Dither (TPDF with xorshift64).

​
Public API

​
Backward-compatible re-exports

Existing imports like songbird_dsp::biquad::BiquadFilter continue to work. New code should prefer the canonical paths (e.g. songbird_dsp::atoms::filter::biquad::BiquadFilter).

​
Conversions (via util)

db_to_gain, gain_to_db, db_to_percent, ballistic_smooth, and variants.

​
Backend

DspBackend trait with RustNativeBackend (pulp SIMD). Operations: biquad_cascade, mix_add, gain, dot, copy_gain, sum, crossfade, clear, peak_abs.

​
How it fits

Foundation crate. Imported by songbird-plugins, songbird-engine, songbird-state, songbird-sync, songbird-headless, songbird-reverb-vst3, songbird-separator. Depends on rustfft and pulp.

​
Roadmap

See ROADMAP.md for the DSP optimization plan covering:
  • Primitive extraction from plugins (OnePole, Phasor, EnvelopeFollower, etc.)
  • Per-sample and block-level optimizations (FMA, scalarization, block-rate coefficients)
  • SIMD strategy (pulp for runtime dispatch, Accelerate/vDSP for macOS)
  • DspNode trait + DspGraph runtime for canvas and vibe-coded plugins

​
Tests

cargo test -p songbird-dsp — runs unit tests, golden-file reference tests (biquad + SVF validated against Python/scipy at 1e-5 tolerance), and backend tests.