Convolver.h File Reference

FFT-based partitioned convolution for long impulse responses. More...

#include "AudioBuffer.h"
#include "AudioSpec.h"
#include "FFT.h"
#include "SimdOps.h"
#include <algorithm>
#include <cassert>
#include <memory>
#include <vector>

Include dependency graph for Convolver.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	dspark::Convolver< T >
	Real-time partitioned convolution using overlap-save with FFT. More...

Namespaces
namespace	dspark
	Main namespace for the DSPark framework.

Detailed Description

FFT-based partitioned convolution for long impulse responses.

Implements the uniform partitioned overlap-save algorithm, which is the standard approach for real-time convolution with long impulse responses (reverb IRs, cabinet simulations, etc.). Breaks the IR into blocks, transforms each block with FFT, and convolves in the frequency domain.

Complexity: O(N log N) per block, vs O(N * M) for direct convolution. For a 2-second reverb IR at 48 kHz (96000 samples), direct convolution would require ~96000 multiplies per sample. FFT-based convolution needs ~20.

Dependencies: FFT.h, AudioBuffer.h, AudioSpec.h, SimdOps.h.

// Load an impulse response from a WAV file:
dspark::WavFile wav;
wav.openRead("cabinet.wav");
auto info = wav.getInfo();
dspark::AudioBuffer<float> ir;
ir.resize(1, static_cast<int>(info.numSamples));
wav.readSamples(ir.toView());
wav.close();
 
// Set up the convolver:
dspark::Convolver<float> conv;
conv.prepare(512, ir.getChannel(0), static_cast<int>(info.numSamples));
 
// In your audio callback:
conv.process(inputBlock, outputBlock, blockSize);

Definition in file Convolver.h.