Windowed-sinc sample rate converter optimized for real-time DSP. More...

#include <Resampler.h>

Public Types
enum class	Quality { Draft , Normal , High , Ultra }

Public Member Functions
void	prepare (double sourceRate, double targetRate, Quality quality=Quality::Normal)
	Prepares the resampler for a given rate conversion.

void	prepare (const AudioSpec &spec, double targetRate, Quality quality=Quality::Normal)
	Prepares the resampler using AudioSpec (unified API).

void	reset () noexcept
	Resets the internal state (delay lines, all channels) to zero.

std::vector< T >	process (const T *input, int inputLength)
	Resamples an entire buffer (offline batch processing).

std::vector< T >	processRange (const T *input, int inputLength, int64_t firstOutput, int64_t count)
	Offline, time-aligned conversion of any span of output samples.

int64_t	getReach () const noexcept
	How far, in output samples, the kernel reaches before an input sample and after it: processRange() over [-getReach(), ceil(inputLength * getRatio()) + getReach()) holds the whole converted signal.

int	processBlock (const T input, int inputLength, T output) noexcept
	Resamples a block of audio in single-channel streaming mode.

int	processBlock (AudioBufferView< T > input, AudioBufferView< T > output) noexcept
	Resamples multi-channel audio using AudioBufferView (streaming).

int	getMaxOutputSamples (int inputLength) const noexcept
	Returns the maximum number of output samples for a given input length.

double	getRatio () const noexcept
	Returns the conversion ratio (targetRate / sourceRate).

int	getFilterLength () const noexcept
	Kernel length in input samples (1 at a ratio of exactly 1).

int	getLatency () const noexcept
	Returns the streaming latency in output samples.

Static Public Attributes
static constexpr int	kMaxTaps = 8192
	Longest kernel, in input samples (see the class note).

Detailed Description

template<typename T>
class dspark::Resampler< T >

Windowed-sinc sample rate converter optimized for real-time DSP.

Quality tiers are filter specifications; the kernel length follows from them and from the ratio (a downsampler's kernel is longer by the ratio, since its band edge is lower relative to the input rate):

Quality	Passband edge	Stopband	Taps, 44.1 <-> 48 kHz
Draft	0.80 Nyquist	60 dB	38 / 40
Normal	0.90 Nyquist	100 dB	130 / 140
High	0.91 Nyquist	140 dB	206 / 224
Ultra	0.915 Nyquist	210 dB	332 / 362

"Nyquist" is the lower of the two rates' Nyquist frequencies; Ultra's passband reaches 20.2 kHz at 44.1 kHz. The float instantiation is limited by float arithmetic to roughly -140 dB; the double one reaches the table. Kernels are capped at kMaxTaps taps, which binds only for downsampling beyond about 25:1 at Ultra: past it the transition band widens.

Template Parameters

T	Sample type (float or double).

Definition at line 85 of file Resampler.h.

Member Enumeration Documentation

◆ Quality

template<typename T >

enum class dspark::Resampler::Quality

strong

Enumerator
Draft	Passband 0.80 of Nyquist, 60 dB stopband: previews.
Normal	Passband 0.90 of Nyquist, 100 dB stopband.
High	Passband 0.91 of Nyquist (20 kHz at 44.1), 140 dB stopband.
Ultra	Passband 0.915 of Nyquist, 210 dB stopband: mastering.

Definition at line 88 of file Resampler.h.

Member Function Documentation

◆ getFilterLength()

template<typename T >

int dspark::Resampler< T >::getFilterLength ( ) const

inlinenoexcept

Kernel length in input samples (1 at a ratio of exactly 1).

Definition at line 315 of file Resampler.h.

◆ getLatency()

template<typename T >

int dspark::Resampler< T >::getLatency ( ) const

inlinenoexcept

Returns the streaming latency in output samples.

The streaming path (processBlock) delays the signal by exactly half the kernel length in input samples – the kernel's group delay – which this getter reports rounded to the nearest output sample. The offline process() path is already time-aligned and has zero latency.

Definition at line 325 of file Resampler.h.

◆ getMaxOutputSamples()

template<typename T >

int dspark::Resampler< T >::getMaxOutputSamples ( int inputLength ) const

inlinenoexcept

Returns the maximum number of output samples for a given input length.

Parameters

inputLength Number of input samples.

Returns: Maximum possible output samples.

Definition at line 303 of file Resampler.h.

◆ getRatio()

template<typename T >

double dspark::Resampler< T >::getRatio ( ) const

inlinenoexcept

Returns the conversion ratio (targetRate / sourceRate).

Definition at line 312 of file Resampler.h.

◆ getReach()

template<typename T >

int64_t dspark::Resampler< T >::getReach ( ) const

inlinenoexcept

How far, in output samples, the kernel reaches before an input sample and after it: processRange() over [-getReach(), ceil(inputLength * getRatio()) + getReach()) holds the whole converted signal.

Definition at line 249 of file Resampler.h.

◆ prepare() [1/2]

template<typename T >

void dspark::Resampler< T >::prepare	(	const AudioSpec &	spec,
		double	targetRate,
		Quality	quality = `Quality::Normal`
	)

inline

Prepares the resampler using AudioSpec (unified API).

Pre-allocates multi-channel states. MUST be called outside the audio thread.

Parameters

spec	Audio environment (sampleRate, numChannels used).
targetRate	Target sample rate in Hz.
quality	Filter specification (default: Normal).

Definition at line 124 of file Resampler.h.

◆ prepare() [2/2]

template<typename T >

void dspark::Resampler< T >::prepare	(	double	sourceRate,
		double	targetRate,
		Quality	quality = `Quality::Normal`
	)

inline

Prepares the resampler for a given rate conversion.

Allocates internal buffers. MUST be called outside the audio thread.

Parameters

sourceRate	Source sample rate in Hz.
targetRate	Target sample rate in Hz.
quality	Filter specification (default: Normal).

Definition at line 108 of file Resampler.h.

◆ process()

template<typename T >

std::vector< T > dspark::Resampler< T >::process	(	const T *	input,
		int	inputLength
	)

inline

Resamples an entire buffer (offline batch processing).

Stateless and time-aligned: output sample k interpolates the input at position k / getRatio() exactly (no latency; the buffer edges are zero-padded). Allocates the output vector – offline use only.

Parameters

input	Source audio samples.
inputLength	Number of input samples.

Returns: Vector of resampled output samples (ceil(inputLength * ratio)).

Definition at line 158 of file Resampler.h.

◆ processBlock() [1/2]

template<typename T >

int dspark::Resampler< T >::processBlock	(	AudioBufferView< T >	input,
		AudioBufferView< T >	output
	)

inlinenoexcept

Resamples multi-channel audio using AudioBufferView (streaming).

Processes each channel sequentially with its independent state.

Parameters

input	Input audio buffer view.
output	Output audio buffer view (pre-allocated).

Returns: Number of output samples produced per channel.

Definition at line 276 of file Resampler.h.

◆ processBlock() [2/2]

template<typename T >

int dspark::Resampler< T >::processBlock	(	const T *	input,
		int	inputLength,
		T *	output
	)

inlinenoexcept

Resamples a block of audio in single-channel streaming mode.

Parameters

input	Input audio samples.
inputLength	Number of input samples.
output	Output buffer (must hold at least getMaxOutputSamples()).

Returns: Number of output samples produced.

Definition at line 262 of file Resampler.h.

◆ processRange()

template<typename T >

std::vector< T > dspark::Resampler< T >::processRange	(	const T *	input,
		int	inputLength,
		int64_t	firstOutput,
		int64_t	count
	)

inline

Offline, time-aligned conversion of any span of output samples.

Output sample k, for k in [firstOutput, firstOutput + count), is the input interpolated at position k / getRatio(), with the input taken as zero outside [0, inputLength). Unlike process(), which stops at the input's last sample, the span may start before 0 and run past the end, so it can hold the kernel's whole response to the signal: the ringing before its first sample and after its last. For an impulse response, whose first sample is often its loudest, that ringing is part of the converted response. getReach() gives the span that holds all of it. Allocates the output vector – offline use only.

Definition at line 208 of file Resampler.h.

◆ reset()

template<typename T >

void dspark::Resampler< T >::reset ( )

inlinenoexcept

Resets the internal state (delay lines, all channels) to zero.

Safe to call from the audio thread after prepare(): it only fills storage that prepare() already sized and never allocates.

Definition at line 140 of file Resampler.h.

Member Data Documentation

◆ kMaxTaps

template<typename T >

constexpr int dspark::Resampler< T >::kMaxTaps = 8192

staticconstexpr

Longest kernel, in input samples (see the class note).

Definition at line 97 of file Resampler.h.

The documentation for this class was generated from the following file:

Resampler.h

Public Types

Public Member Functions

Static Public Attributes

Detailed Description

Member Enumeration Documentation

◆ Quality

Member Function Documentation

◆ getFilterLength()

◆ getLatency()

◆ getMaxOutputSamples()

◆ getRatio()

◆ getReach()

◆ prepare() [1/2]

◆ prepare() [2/2]

◆ process()

◆ processBlock() [1/2]

◆ processBlock() [2/2]

◆ processRange()

◆ reset()

Member Data Documentation

◆ kMaxTaps