shortwavlabs/agent-skills

dsp

Digital signal processing algorithms and techniques for real-time audio.

Voir la source

Mode d’emploi

  1. Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
  2. Consultez le document original pour les outils, identifiants et environnements requis.
  3. Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/shortwavlabs/agent-skills --skill dsp

Avant de commencer

  • L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
  • Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Document Skill original

Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.

DSP — Digital Signal Processing for Audio

A comprehensive reference for implementing real-time audio DSP algorithms in C/C++. Derived from the musicdsp.org community archive covering two decades of practitioner knowledge.

When to Use This Skill

  • Implementing audio filters, effects, synthesizers, or analyzers in C/C++
  • Writing processBlock(), process(), or sample-by-sample DSP loops
  • Designing biquad coefficients, Moog ladder filters, or state variable filters
  • Building delay lines, reverbs, compressors, waveshapers, or stereo processors
  • Generating bandlimited waveforms, noise, or envelopes
  • Optimizing audio code with fast math approximations or denormal prevention

For circuit-accurate emulation of a specific analog circuit (matching a schematic or SPICE reference), load circuit-to-dsp as well: it decides structure, solver, integrator and rate from measured error; this skill supplies the building blocks.

Architecture

The skill is organized into domain-specific references. Read the relevant file(s) for the task at hand:

ReferenceCoverageWhen to read
references/filters.mdBiquad/RBJ, Moog ladder, SVF, FIR, DC block, formant, parameter smoothing, crossoverAny filter implementation
references/effects.mdDelay, reverb, compressor/limiter, phaser/wah, waveshaper/distortion, lo-fi, stereo, convolutionAny audio effect
references/synthesis.mdOscillators, bandlimited waveforms, FM/PM, DSF, noise, envelopes, Chebyshev waveshaping, LFOsAny sound generation
references/analysis.mdFFT, DFT, Goertzel, Walsh-Hadamard, envelope following, beat detection, LPC, limitingSpectral analysis, detection
references/utilities.mdFast math (exp, log, sin, sqrt), interpolation, clipping, denormals, dithering, MIDI conversion, parameter mapping, benchmarking, lock-free FIFOPerformance optimization, conversion, audio safety

Implementation Principles

These principles apply across all domains:

Sample Rate Independence

Always derive coefficients from the sample rate. Never hard-code values that assume 44.1 kHz.

cpp
// Coefficient from frequency and sample rate
float w0 = 2.0f * M_PI * cutoffHz / sampleRate;

Denormal Prevention

Floating-point denormals (subnormal numbers near zero) cause massive CPU stalls on x86. Add a tiny offset to state variables:

cpp
static constexpr float kDenormal = 1e-24f;
// In process loop:
state += kDenormal;

Alternatively, flush-to-zero compiler flags (-ffast-math on GCC/Clang) handle this at the CPU level.

Parameter Smoothing

Never change filter coefficients or gain values instantaneously — this causes clicks. Smooth parameters with a one-pole lowpass:

cpp
// Time-based smoothing coefficient
float a = expf(-2.0f * M_PI / (smoothTimeMs * 0.001f * sampleRate));
// Per sample:
smoothedValue = a * smoothedValue + (1.0f - a) * targetValue;

Audio Thread Safety

The process callback must never allocate memory, take locks, or call blocking OS functions. Pre-allocate all buffers, use lock-free communication for parameter updates, and avoid virtual dispatch in hot loops.

Bilinear Transform / Prewarping

Most digital filters are derived from analog prototypes via the bilinear transform. The key mapping:

w0 = 2 * pi * fc / Fs

For higher accuracy at high frequencies, apply frequency prewarping:

wa = 2 * Fs * tan(w0 / 2)

Code Conventions

All code examples in this skill use C/C++ with these conventions:

  • float for audio samples (32-bit), double for coefficient calculation
  • sampleRate or sr for sample rate in Hz
  • M_PI for pi (define as #ifndef M_PI ... #endif for portability)
  • Processing functions take input sample and return output sample
  • State variables are class members (not globals)
  • All frequencies in Hz unless explicitly noted otherwise