GCCV7
Overview
Compute angular 2D direction of arrival of audio sources arriving at the mic array
Discussion
This module computes angular 2D direction of arrival of audio sources arriving at the mic array
Input Pins:
Input Pin 1: Multichannel frequency domain data. Usually output of a WOLA Analysis Module
Number of channels must match the number of microphones in the micGeometry argument
Output Pins:
Output Pin 1: Estimated angular direction of arrival for this WOLA block in degrees. This will be an integer between 0 and 360.
Output Pin 2: Confidence value for this angle estimate. This will be an integer. Larger values mean more confidence.
Output Pin 3: Array of 360 floats denoting a histogram of possible angles of arrival for this block. Output pin 1 will be the peak of this histogram.
Module Arguments:
numMics: number of microphones being used
interpFactor: Interpolation factor for better angular resolution at the cost of more CPU resources
fftSize: FFTSize used by WOLA Analysis block preceding the SbDOAV2 module
Fs: Sampling rate of time domain data entering WOLA Analysis block
LowerBin: Frequency bin index of lowest WOLA bin entering SbDOAV2 input pin
Inspector Tuning Variables:
BlocksPerHistogram: Each WOLA block we get an angle estimate for each bin. These are accumulated over BlocksPerHistogram
WOLA blocks into a histogram. Then the peak of the histogram is output as the angle estimate. Larger values of
BlocksPerHistogram give more accuracy but slower response time. BlocksPerHistogram value does not affect MIPS.
smoothingCoefficient: The smoothing coefficient, (SC), is a float which takes on values between 0 and 1.
SC operates on the relevant portion of the time domain correlation data between every mic pair. SC performs
exponential smoothing/averaging in time over each point of the correlation functions. The equation is:
state(n) = SC*(new data) + (1-SC)*state(n-1). This smoothing increases the accuracy of the DOA algorithm.
SC allows us to trade off accuracy and responsiveness. The smaller the value of SC, the longer the tail
of our exponential smoothing/averaging.
SC smaller ---> more smoothing more accuracy but less responsiveness
SC larger ---> less smoothing more responsiveness but less accuracy
Type Definition
typedef struct _ModuleGCCV7
{
ModuleInstanceDescriptor instance; // Common Audio Weaver module instance structure
INT32 numMics; // Number of mics in the array
INT32 interpFactor; // Interpolation factor
INT32 fftSize; // Size of FFT upstream from this module
INT32 MultiSource; // Select single or multiple audio sources reported per block
FLOAT32 Fs; // Time domain sampling rate before WOLA/FFT
INT32 LowerBin; // Frequ bin index of lowest frequ component in the WOLA block
FLOAT32 smoothingCoefficient; // Smoothing coefficient for exponential smoothing of frequ domain CC data
INT32 BlocksPerHistogram; // Number of blocks used to fill histogram before doing smoothing and averaging
INT32 interpBS; // Block Size at interpolated rate
FLOAT32 interpSR; // Sample rate at interpolated rate
INT32 numDirs; // Number of look directions
INT32 maxNumSources; // Maximum number of sources to identify
INT32 useInputWeightingPin; // Use input weighting pin
INT32 smoothing; // 0 No Smoothing 1 Smoothing
INT32 useSmoothingPin; // Select use control pin for smoothing coefficient
INT32 combineCongruentMicPairs; // Select combining congruent mic pairs for MIPS savings
INT32 boxcarDelay; // Half length of rectangular window for histogram smoothing
INT32 gaussDelay; // Half length of gaussian window for histogram smoothing
FLOAT32 gaussSigma; // Standard Deviation for Gaussian window
INT32 NumHistogramStates; // Number of full histograms used for temporal averaging
INT32 QCOpt; // Select optimization. DONT CARE, NOT IMPLEMENTED YET
INT32 doHistogramProcessing; // Add histogram processing to module outputs
INT32 numBins; // Number of bins. This should be the input block size
INT32 numPairs; // Number of microphone pairs
INT32 xCorrsZeroSamp; // 0th sample index in the xcorr data
INT32 statesSize; // Size of states array for frequency domain smoothing
INT32 tdCurrSize; // Size of temp td array for time domain smoothing
INT32 maxDelay; // Maximum of gaussDelay, boxcarDelay
INT32 HistStatePtr; // Index of oldest entry in histogram buffer
INT32 blockCtr; // Counter to keep track of number of blocks used to update histogram
INT32 blockCtrReset; // Counter to keep track of number of blocks used between reset to zero of avgHistogram
INT32 blockCtrMultiOpt; // Counter for multisource/multiblock optimization
INT32 numPairsToUse; // Number of non-redunbdant microphone pairs
INT32 NSRMP; // Number of sets of redundant mic pairs
INT32 maxNMPiS; // max number of redundant pairs in all sets of redundant pairs
INT32 NNRMP; // Number of non-redundant mic pairs
FLOAT32 epsilon; // Small number
FLOAT32 c; // Speed of sound in meters per sec
FLOAT32 windowSize; // How much to extend the look window when comparing measured delays with theoretical time delays. This is relative number.
INT32 removalWindowDelta; // Number of samples to remove when filtering DOA Peaks
INT32 doHistSmoothing; // Specifies whether to do smoothing of histogram
INT32 doHistAveraging; // Specifies whether to do temporal smoothing of histogram
FLOAT32 MinPeakThresh; // Minimum height of a histogram peak to be eligible to be reported as a DOA
INT32 MinDist; // Distance in integer units of degrees, between peaks in histogram to count as a distinct DOA
FLOAT32 noiseFloorDBOffset; // Offset, in DB, for noise floor weighting. Higher value will weight noise floor less.
INT32 lookDirLow; // Min look direction, in degrees. The low end of the DOA look direction.
INT32 lookDirHigh; // Max look direction, in degrees. The high end of the DOA look direction.
INT32 tauMaxMax; // Maximum theoretical time delays for all mic pair
INT32 tauMinMin; // Minimum theoretical time delays for all mic pair
INT32 tdZeroSamp; // windowed xcorr 0 delay sample index
INT32 tdLength; // Length of the xcorr vector, after windowing
FLOAT32* micArrayCoords; // Microphone geometry in cartesian coordinates and in meters. The vector going from (0,0) to (1,0) is considered 0 degrees.
FLOAT32* noiseFloorVar; // Microphone noise floor measurement, in db10 units.
FLOAT32* HistEnergies; // Holds energy of sources at each source direction for a data block
FLOAT32* HistAngles; // Holds angle for each source for a data block
FLOAT32* HistState; // Holds NumHistogramStates histograms
FLOAT32* avgHistogram; // Scratch array to hold average histogram
FLOAT32* currentHistogram; // Scratch array to hold current histogram
FLOAT32* smoothedHistogram; // Scratch array to hold smoothed histogram
FLOAT32* histExt; // Scratch array to hold smoothed histogram
FLOAT32* histCnv; // Scratch array to hold smoothed histogram
INT32* histSortedAngles; // Scratch array for histogram peak picking
FLOAT32* histSortedValues; // Scratch array for histogram peak picking
INT32* histAnglesUsed; // Scratch array for histogram peak picking
FLOAT32* valuesOutState; // Scratch array to hold output histogram counts
INT32* anglesOutState; // Scratch array to hold output angles
FLOAT32* GaussianWindow; // Gaussian window for smoothing histogram
FLOAT32* BoxcarWindow; // Rectangular window for smoothing histogram
INT32* taus; // Theoretical time delays based on the mic geometry and other factors.
FLOAT32* deltaTaus; // Difference between taus and taus rounded to nearest sample delay.
FLOAT32* lookDirs; // Look directions, in degrees.
INT32* tauMax; // Maximum theoretical time delays for each mic pair
INT32* tauMin; // Minimum theoretical time delays for each mic pair
FLOAT32* doaValues; // Temporary variable from the processing function
FLOAT32* td; // Temporary variable from the processing function
INT32* Map; // Map from pair processed to index of pair in micPairsNR = array of non-redundant mic pairs
FLOAT32* states; // Holds current estimate of Rxx
FLOAT32* tdCurr; // Temp array for new td data when doing time domain smoothing
INT32* NMPiS; // Array containing number of redundant mic pairs in each set
INT32* CIR; // Array containing channel indices of redundant mic pairs
INT32* CINR; // Array containing channel indices of non-redundant mic pairs
void * ifft_struct_pointer; // Points to an instance of an IFFT module
} ModuleGCCV7Class;
Variables
Properties
Name | Type | Usage | isHidden | Default value | Range | Units |
numMics | int | const | 0 | 4 | Unrestricted | |
interpFactor | int | const | 0 | 4 | Unrestricted | |
fftSize | int | const | 0 | 256 | Unrestricted | |
MultiSource | int | const | 0 | 0 | Unrestricted | |
Fs | float | const | 0 | 16000 | Unrestricted | |
LowerBin | int | const | 0 | 0 | Unrestricted | |
smoothingCoefficient | float | parameter | 0 | 0.2 | 0:1 | |
BlocksPerHistogram | int | parameter | 0 | 32 | 1:200 | |
interpBS | int | derived | 0 | 512 | Unrestricted | |
interpSR | float | derived | 0 | 64000 | Unrestricted | |
numDirs | int | const | 1 | 360 | Unrestricted | |
maxNumSources | int | const | 1 | 1 | Unrestricted | |
useInputWeightingPin | int | const | 1 | 0 | Unrestricted | |
smoothing | int | const | 1 | 2 | Unrestricted | |
useSmoothingPin | int | const | 1 | 0 | Unrestricted | |
combineCongruentMicPairs | int | const | 1 | 1 | Unrestricted | |
boxcarDelay | int | const | 1 | 2 | Unrestricted | |
gaussDelay | int | const | 1 | 8 | Unrestricted | |
gaussSigma | float | const | 1 | 3.2 | Unrestricted | |
NumHistogramStates | int | const | 1 | 4 | Unrestricted | |
QCOpt | int | const | 1 | 1 | Unrestricted | |
doHistogramProcessing | int | const | 1 | 1 | Unrestricted | |
numBins | int | const | 1 | 129 | Unrestricted | |
numPairs | int | const | 1 | 6 | Unrestricted | |
xCorrsZeroSamp | int | const | 1 | 64 | Unrestricted | |
statesSize | int | const | 1 | 1 | Unrestricted | |
tdCurrSize | int | const | 1 | 1 | Unrestricted | |
maxDelay | int | const | 1 | 8 | Unrestricted | |
HistStatePtr | int | state | 1 | 0 | Unrestricted | |
blockCtr | int | state | 1 | 0 | Unrestricted | |
blockCtrReset | int | state | 1 | 0 | Unrestricted | |
blockCtrMultiOpt | int | state | 1 | 0 | Unrestricted | |
numPairsToUse | int | const | 1 | 6 | Unrestricted | |
NSRMP | int | const | 1 | 1 | Unrestricted | |
maxNMPiS | int | const | 1 | 1 | Unrestricted | |
NNRMP | int | const | 1 | 1 | Unrestricted | |
epsilon | float | parameter | 1 | 1e-16 | Unrestricted | |
c | float | parameter | 1 | 343 | Unrestricted | |
windowSize | float | parameter | 1 | 0 | Unrestricted | |
removalWindowDelta | int | parameter | 1 | 2 | Unrestricted | |
doHistSmoothing | int | parameter | 1 | 1 | Unrestricted | |
doHistAveraging | int | parameter | 1 | 1 | Unrestricted | |
MinPeakThresh | float | parameter | 1 | 1 | 0:100 | |
MinDist | int | parameter | 1 | 10 | 0:360 | degrees |
noiseFloorDBOffset | float | parameter | 1 | 5 | -20:20 | |
lookDirLow | int | parameter | 1 | 0 | 0:360 | |
lookDirHigh | int | parameter | 1 | 360 | 0:360 | |
tauMaxMax | int | derived | 1 | 0 | Unrestricted | |
tauMinMin | int | derived | 1 | 0 | Unrestricted | |
tdZeroSamp | int | derived | 1 | 1 | Unrestricted | |
tdLength | int | derived | 1 | 1 | Unrestricted | |
micArrayCoords | float* | parameter | 0 | [4 x 2] | Unrestricted | |
noiseFloorVar | float* | parameter | 1 | [129 x 1] | -500:500 | |
HistEnergies | float* | state | 0 | [1 x 1] | Unrestricted | |
HistAngles | float* | state | 0 | [1 x 1] | Unrestricted | |
HistState | float* | state | 0 | [1440 x 1] | Unrestricted | |
avgHistogram | float* | state | 0 | [360 x 1] | Unrestricted | |
currentHistogram | float* | state | 0 | [360 x 1] | Unrestricted | |
smoothedHistogram | float* | state | 0 | [360 x 1] | Unrestricted | |
histExt | float* | state | 0 | [376 x 1] | Unrestricted | |
histCnv | float* | state | 0 | [392 x 1] | Unrestricted | |
histSortedAngles | int* | state | 0 | [360 x 1] | Unrestricted | |
histSortedValues | float* | state | 0 | [360 x 1] | Unrestricted | |
histAnglesUsed | int* | state | 0 | [360 x 1] | Unrestricted | |
valuesOutState | float* | state | 0 | [10 x 1] | Unrestricted | |
anglesOutState | int* | state | 0 | [10 x 1] | Unrestricted | |
GaussianWindow | float* | parameter | 1 | [17 x 1] | Unrestricted | |
BoxcarWindow | float* | parameter | 1 | [5 x 1] | Unrestricted | |
taus | int* | derived | 1 | [6 x 360] | Unrestricted | |
deltaTaus | float* | derived | 1 | [6 x 360] | Unrestricted | |
lookDirs | float* | derived | 1 | [1 x 360] | Unrestricted | |
tauMax | int* | derived | 1 | [6 x 1] | Unrestricted | |
tauMin | int* | derived | 1 | [6 x 1] | Unrestricted | |
doaValues | float* | state | 0 | [360 x 1] | Unrestricted | |
td | float* | state | 0 | [1 x 6] | Unrestricted | |
Map | int* | derived | 0 | [6 x 1] | Unrestricted | |
states | float* | state | 0 | [1 x 1] | Unrestricted | |
tdCurr | float* | state | 0 | [1 x 1] | Unrestricted | |
NMPiS | int* | derived | 0 | [100 x 1] | Unrestricted | |
CIR | int* | derived | 0 | [100 x 1] | Unrestricted | |
CINR | int* | derived | 0 | [100 x 1] | Unrestricted | |
ifft_struct_pointer | void * | state | 1 | Unrestricted |
Pins
Input Pins
Name: micIn
Description: Microphone array input - in frequency domain
Data type: float
Channel range: 4
Block size range: Unrestricted
Sample rate range: Unrestricted
Complex support: Complex
Output Pins
Name: estDirs
Description: Estimated directions, in degrees
Data type: float
Name: estEnergy
Description: Estimated relative energy for each estimated direction
Data type: float
Name: HistValues
Description: Histogram output values
Data type: float
Scratch Pins
Channel count: 1
Block size: 513
Sample rate: 48000
Channel count: 1
Block size: 1024
Sample rate: 48000
Channel count: 1
Block size: 1024
Sample rate: 48000
MATLAB Usage
File Name: gcc_v7_module.m
M = gcc_v7_module(NAME, numMics, interpFactor, fftSize, FS, LOWERBIN)
GCCV7 module identifies the 2D angular direction of an audio source received at the mic array.
The module takes multichannel frequency domain data as input, (usually the output of a WOLA
Analysis module). The module produces 2 outputs per block:
1.)Angle = estimated angular direction of arrival in degrees. This will be an integer between 0 and 360.
2.)Confidence value for this angle estimate. Larger values mean more confidence in the angle estimate.
Arguments:
NAME - name of the module.
numMics - number of microphones in the microphone array
interpFactor - how much to interpolate for better estimation at the
cost of more MIPS
fftSize - Size of the FFT operation done upstream from this module.
FS - Sampling rate of time domain data entering WOLA Analysis block
LOWERBIN - Frequency bin index of lowest WOLA bin entering GCCV7 input pin