Document toolboxDocument toolbox

(8.D.2.3) SCNRv4GainCalculator

Owned by Catherine Spain
Last updated: Jan 11, 2024 by Andrew Perrin
3 min read

Overview

DSP Concepts proprietary Single-Channel Noise Reduction Gain Calculator. For version 4

Discussion

This module calculates gain coefficients for single-channel noise reduction. It is part of the SCNRv4 design.

The module works in conjunction with the WOLA filterbank modules.

The Yk input pin expects the squared-magnitude of the WOLA-domain noisy signal.

The Pn input pin expects a signal with noise levels per subband. This signal should be in dBs, and typically comes directly from the output of the NoiseTracking module.

The output pin provides the gain coefficients per subband, which can then be applied to the WOLA-domain samples of the input signal. These are real numbers and represent magnitude only.

The module is single channel, expects real data, and operates on any block size and any sampling rate.

The minGaindB parameter is the lower clipping threshold of gain for each frequency bin. The value for minGaindB is a negative number in dB, and it corresponds to the maximum attenuation allowed for each subband.

Before being output, the gain signal is smoothed out using cross-bin averaging in logarithmic groupings. The smoothOctave parameter defines how many octaves the averaging should span. Smoothing can be disabled by setting smoothOctave to 0.

The timeAlphaPrioriMs parameter controls the recursive averaging parameter in the SNR calculation. The default value is 250ms. Higher values mean more smoothing, resulting in a lower risk of musical noise, but a higher risk of distorting onset transients.

Type Definition

typedef struct _ModuleSCNRv4GainCalculator { ModuleInstanceDescriptor instance; // Common Audio Weaver module instance structure FLOAT32 prioriSnrScaler1Gain; // Scaler gain. FLOAT32 prioriSnrScaler2Gain; // Scaler gain. FLOAT32 timeAlphaPrioriMs; // Averaging time for a-priori SNR computation, in ms. FLOAT32 minGaindB; // Minimum gain (maximum reduction) per subband, in dB. FLOAT32 smoothOctave; // Number of octaves for cross-bin logarithmic smoothing. FLOAT32 postSnrClipAsymUpper; // Asymmetric-clip upper limit. FLOAT32 prioriSnrClipAsymUpper; // Asymmetric-clip upper limit. FLOAT32 gainCalcLogScaleGain; // Scale Offset gain. FLOAT32 smoothStartFreq; // Start frequency for cross-bin smoothing. FLOAT32* aAPrioriFb; // A-priori SNR estimation feedback line. FLOAT32* aPrioriSnrDelay; // Block Delay single-block delay buffer. INT32* aSmoothStartIdx; // Starting index for the averaging process. INT32* aSmoothEndIdx; // Ending index for the averaging process. FLOAT32* aSmoothOneOverN; // Precomputed 1/N for averaging. } ModuleSCNRv4GainCalculatorClass;

Variables

Properties

Name

Type

Usage

isHidden

Default value

Range

Units

prioriSnrScaler1Gain

float

derived

0

0.99992

Unrestricted

 

prioriSnrScaler2Gain

float

derived

0

8.3327e-05

Unrestricted

 

timeAlphaPrioriMs

float

parameter

0

250

0:1000000

ms

minGaindB

float

parameter

0

-12

-30:0

dB

smoothOctave

float

parameter

0

0.1

0:1

 

postSnrClipAsymUpper

float

const

1

1000000

Unrestricted

 

prioriSnrClipAsymUpper

float

const

1

10000

Unrestricted

 

gainCalcLogScaleGain

float

const

1

0.033333

Unrestricted

 

smoothStartFreq

float

const

1

0

Unrestricted

 

aAPrioriFb

float*

state

1

[1 x 33]

Unrestricted

 

aPrioriSnrDelay

float*

state

1

[1 x 33]

Unrestricted

 

aSmoothStartIdx

int*

derived

1

[33 x 1]

Unrestricted

 

aSmoothEndIdx

int*

derived

1

[33 x 1]

Unrestricted

 

aSmoothOneOverN

float*

derived

1

[33 x 1]

Unrestricted

 

Pins

Input Pins

Name: Yk

Description: Input signal power

Data type: float

Channel range: 1

Block size range: Unrestricted

Sample rate range: Unrestricted

Complex support: Real

 

Name: Pn

Description: Input noise power

Data type: float

Channel range: 1

Block size range: Unrestricted

Sample rate range: Unrestricted

Complex support: Real

Output Pins

Name: out

Description: Output signal

Data type: float

Scratch Pins

Channel count: 1

Block size: 33

Sample rate: 48000

MATLAB Usage

File Name: scnrv4_gain_calculator_module.m

M = scnrv4_gain_calculator_module(NAME) Single-channel noise-reduction gain calculator. This module takes a noise-level estimate and calculates gain coefficients to be used in noise reduction. Arguments: NAME - name of the module

Â