DSPark
](LICENSE)
DSPark is a header-only C++20 audio DSP library for developers building real-time and offline audio software. It provides effects, filters, synthesis and spectral building blocks, analysis and musical tools, native file I/O, and an optional VST3/CLAP/Audio Unit v2 plugin layer.
The framework modules have no third-party dependencies and require no separately compiled DSPark library. Optimized paths may use compiler-provided SIMD intrinsics behind platform guards. The plugin adapters and the Windows-only DSParkLab application are separate from the umbrella header and carry their own platform integration requirements.
The v1.8.0 CI matrix covers Windows, Linux, macOS, WebAssembly, and an embedded profile. The framework headers also target iOS and Android, although those two platforms do not have dedicated v1.8.0 CI jobs.
Current release: v1.8.0 | API reference | Cookbook | Examples | Plugin guide
Quick start
Requirements
- A C++20 compiler.
- CMake 3.21 or newer when using the supplied CMake project.
- An optimized build for production DSP (
-O2or/O2); header-only code is compiled with the flags of the consuming target.
Add DSPark to a project
Vendor the repository at third_party/DSPark, add that directory to your include path, and include the umbrella header:
For a direct compiler invocation:
For CMake, the interface target supplies the include path and C++20 requirement:
A CMake installation is consumed with find_package(dspark CONFIG REQUIRED) and the same dspark::dspark target. Every route (subdirectory, FetchContent, installed package) also provides the plugin layer and the dspark_add_plugin() bundle helper described in docs/plugins.md.
Or fetch the v1.8.0 release tag:
Prepare and process a block
A typical block processor is constructed and configured with prepare(AudioSpec) before its audio-stream owner calls processBlock().
In an audio host, create an AudioBufferView<float> over the host-owned channel pointers instead of allocating an AudioBuffer in the callback. See the real-time-style channel-strip example and the offline WAV example for complete programs.
Capabilities
| Area | Representative APIs |
|---|---|
| Filters and equalization | Biquad, StateVariableFilter, LadderFilter, FilterEngine, Equalizer, CrossoverFilter |
| Dynamics and level control | Compressor, Limiter, NoiseGate, Expander, DynamicEQ, MultibandCompressor, DeEsser |
| Offline dynamics, clipping, stereo and analysis | OfflineLeveler, OfflineBeatCompressor, OfflinePeakCompressor, OfflinePunch, OfflineSoftClipper, OfflineHardClipper, OfflineStereoGenerator, OfflineStereoBalance, OfflineEnergyAnalyzer, OfflineTransientAnalyzer, OfflineTempoAnalyzer (worker API) |
| Nonlinear and analog modeling | Saturation, Clipper, TapeMachine, TubePreamp, TransformerModel, wdf::* |
| Time, pitch, and spatial processing | Delay, AlgorithmicReverb, Reverb, PitchShifter, TimeStretch, PitchCorrector, GranularProcessor |
| Spectral processing | FFTReal, Convolver, ZeroLatencyConvolver, SpectralProcessor, SpectralFreeze, SpectralDenoiser |
| Analysis and metering | SpectrumAnalyzer, LoudnessMeter, AudioIntervalAnalyzer, LoudnessNormalizer, PitchDetector, OnsetDetector, BeatTracker, LoopFinder, DelayEstimator |
| Synthesis and music | Oscillator, WavetableOscillator, Sampler, ADSREnvelope, ChordDetector, KeyDetector, harmony::* |
| Infrastructure | AudioBuffer, ProcessorChain, Resampler, Oversampling, Dither, SpscQueue, versioned state blobs |
| File I/O | WAV and MP3 read/write, Standard MIDI File read/write, native FLAC decode |
| Plugin development | Native VST3 and CLAP adapters, macOS Audio Unit v2, instruments, sidechains, automation, state, and optional WebView editors |
The API reference is the exhaustive class and method inventory. The cookbook covers production workflows including mastering, restoration, tempo detection, time stretching, loop finding, and pitch correction.
File I/O qualification
WAV and MP3 support includes reading and writing. MidiFile reads Standard MIDI File formats 0, 1, and 2 and writes formats 0 and 1. FlacFile is a decode-only native FLAC implementation; its write operations return false. Define DSPARK_NO_FILE_IO before including DSPark.h to omit all file-I/O headers for targets without filesystem support.
Real-time and threading contract
Real-time safety is a per-method contract, not a blanket property of every operation:
- Construction,
prepare(), state restoration, and setup-only configuration may allocate and must not run concurrently with processing. - A prepared processor has one audio-stream owner. Processing methods that a component documents as real-time-safe avoid allocation and unbounded waits; offline utilities are outside that contract.
- Parameter publication normally assumes one control writer and one audio owner. Some setters are safe during processing; setup setters such as changes that resize oversampling storage are not.
- Before sharing an instance, read its documented threading contract. Report the latency of lookahead, FFT, convolution, and oversampled paths to the host.
The complete ownership and publication rules are in the threading model.
Platform and validation coverage
| Target | v1.8.0 validation |
|---|---|
| Windows | MSVC x64 and ARM64: test and conformance suites |
| Linux | GCC x64/ARM64 and Clang x64: test and conformance suites |
| macOS | Clang ARM64: test and conformance suites |
| WebAssembly | Emscripten conformance in scalar and SIMD128 configurations |
| Embedded profile | GCC compile gate with file I/O, exceptions, and RTTI disabled |
| iOS and Android | Portable framework headers; not separate targets in the v1.8.0 CI matrix |
CI also checks installed-header self-sufficiency with GCC and Clang, warning-free builds, sanitizers, concurrent publication paths, examples, native plugin smoke hosts and validators, and generated documentation. The official EBU loudness-vector job is best effort because the licensed test material is downloaded separately; the job reports a skip when the vectors are unavailable.
Measured processor and analyzer results, including settings and limitations, are published in the quality metrics table.
Native plugins
DSPark supplies native adapters for VST3 and CLAP on desktop platforms and Audio Unit v2 on macOS. The required VST3 C API and CLAP headers are included in the repository; the CMake helper builds the platform-specific module or bundle layout.
The shared plugin contract covers parameters, automation, state, latency, tail reporting, bypass, mono/stereo layouts, sidechains, transport, MIDI instruments, offline-rendering modes, and factory presets. Custom editors can be written in HTML/CSS/JavaScript and embedded with the platform WebView layer. On Linux, a custom editor requires WebKitGTK at runtime; the host's generic parameter UI remains the fallback when a WebView is unavailable.
Start with the plugin guide and the plugin examples.
DSParkLab
DSParkLab/ is a Windows-only Win32/Direct3D 11 application for interactive processor testing. Its bundled Dear ImGui and miniaudio sources are confined to the application and are not dependencies of the DSPark framework. Build it with DSParkLab\build.bat (Visual Studio 2019 or later, any edition).
Documentation
- API reference - generated from the public headers and guides.
- Cookbook - task-oriented DSP recipes.
- Examples - standalone processing and plugin projects.
- Plugin guide - formats, host contract, editors, and shipping checks.
- Threading model - setup, control, audio, and readout ownership.
- Quality metrics - measured settings, results, and caveats.
- v1.8.0 migration guide - changed defaults, plugin timing and packaging from v1.7.0.
- v1.7.0 migration guide - source and latency changes from v1.6.1.
- Changelog - release history.
What's new in v1.8.0
- Complete-source offline analysis and automatic Leveler, Peak/Beat compression, Punch, soft/hard clipping, stereo generation and stereo balance. Immutable plans, exact exclusions, source validation and bounded worker storage are documented with measured quality and resource costs.
StereoGeneratoralso supports real-time stereo processing, preserving the original mid through processed-copy delta add-back. It defaults to 1x; higher oversampling factors remain explicit options.AudioIntervalAnalyzermeasures exact sample intervals from blocks without retaining PCM, with explicit validity and independent/continuous loudness context. See the interval contract.TubePreampintegrates the nonlinear circuit within each internal sample at 2x and above, retaining the 2x default and configurable 1x operation. The documented tone sweeps include +36 dB drive; stage changes use a primed crossfade without callback allocations.- Rational tempo alternatives in
BeatTracker, a polyphonic multi-zoneSampler, Studio time/pitch processing, and conversion-specificResamplerfilters with explicit latency and finite-source range conversion. AlgorithmicReverbrebuilt as a true-stereo 32-line FDN (16 in Eco) with a binaural early field and frequency-dependent decay control.- A split-format Stockham FFT and a polyphase
Oversamplingdecimator, with SIMD paths and benchmark conditions documented in their headers. - Band-limited oscillators, analog-matched EQ bells, a Wiener-gain
SpectralDenoiser, gapless MP3 round trips, and glided parameter changes across the effects. - Plugin layer: latency-aligned bypass,
choice()/stepped()parameters, UI-thread restart requests, chunked oversize blocks, and an installed CMake package that ships the plugin layer withdspark_add_plugin().
Read the release notes, full changelog, and migration guide.
Build and contribute
From a repository checkout:
Contributions are welcome. Read CONTRIBUTING.md for the development workflow and conventions. Please open an issue to discuss a proposed change before submitting a pull request. Bug reports should include the platform, compiler version, and a minimal reproduction.
License
DSPark is available under the [MIT License](LICENSE).
Author
Created and maintained by Cristian Moresi. Contact: dev@cristianmoresi.com.
