Dither and noise shaping for bit depth reduction. More...

#include <gin_dither.h>

Public Types
enum	NoiseType { none = 0 , rectangular , triangular , gaussian }
	Probability distribution of the dither noise. More...

enum	Shaping { flat = 0 , highpass }
	Spectrum of the dither noise itself. More...

Public Member Functions
	Dither ()=default

	~Dither ()=default

void	setBitDepth (int bits)
	Target bit depth.

void	setNoiseType (NoiseType t)

void	setShaping (Shaping s)

void	setAmount (float lsbs)
	Dither depth in LSBs.

void	setNoiseShaping (float ns)
	Error feedback noise shaping, 0 to 1.

void	reset ()

void	process (juce::AudioSampleBuffer &buffer)

Detailed Description

Dither and noise shaping for bit depth reduction.

Truncating to a lower bit depth correlates the error with the signal, which is what makes quiet fades sound grainy and reverb tails turn to gravel rather than simply getting quieter. Adding a small amount of noise before quantising decorrelates it: the error becomes hiss, which is far less objectionable than distortion that tracks the music.

Noise shaping goes further and feeds the quantisation error back through a second order filter, pushing the noise out of the midrange where the ear is most sensitive.

Key Features:

Rectangular, triangular and near-gaussian noise
Optional highpass shaped noise
Second order error feedback noise shaping
Any bit depth

Usage:

Dither dither;
dither.setBitDepth (16);
dither.setNoiseType (Dither::triangular);   // the usual choice
dither.setNoiseShaping (0.5f);
 
dither.process (buffer);

Note: TPDF (triangular) at 2 LSB is the standard, safe default - it fully decorrelates the error at the cost of a little more noise than RPDF.

See also: BitCrusher

Member Enumeration Documentation

◆ NoiseType

enum Dither::NoiseType

Probability distribution of the dither noise.

none: no dither, straight truncation - included for comparison
rectangular: RPDF, one random value. Quietest, but the error is not fully decorrelated and the noise floor still moves with the signal
triangular: TPDF, sum of two. The standard choice
gaussian: sum of ten, approaching a normal distribution

Enumerator
none
rectangular
triangular
gaussian

◆ Shaping

enum Dither::Shaping

Spectrum of the dither noise itself.

flat: white
highpass: the difference of successive values, tilting the noise up and away from where the ear is most sensitive

Enumerator
flat
highpass

Constructor & Destructor Documentation

◆ Dither()

Dither::Dither ( )

default

◆ ~Dither()

Dither::~Dither ( )

default

Member Function Documentation

◆ setBitDepth()

void Dither::setBitDepth ( int bits )

Target bit depth.

16 for CD, 8 for something lo-fi on purpose.

◆ setNoiseType()

void Dither::setNoiseType ( NoiseType t )

◆ setShaping()

void Dither::setShaping ( Shaping s )

◆ setAmount()

void Dither::setAmount ( float lsbs )

Dither depth in LSBs.

2 is the usual amount for TPDF.

◆ setNoiseShaping()

void Dither::setNoiseShaping ( float ns )

Error feedback noise shaping, 0 to 1.

◆ reset()

void Dither::reset ( )

◆ process()

void Dither::process ( juce::AudioSampleBuffer & buffer )

References highpass.

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

gin_dither.h

Public Types

Public Member Functions