Code covered by the BSD License  

Highlights from
Precedence-based cross-correlograms

Be the first to rate this file! 5 Downloads (last 30 days) File Size: 217 KB File ID: #27776 Version: 1.16

Precedence-based cross-correlograms

by

 

27 May 2010 (Updated )

Calculate cross-correlograms with a wide range of options

| Watch this File

File Information
Description

Note:- Requires a C compiler and the following function: http://www.mathworks.co.uk/matlabcentral/fileexchange/27140-check-whether-mex-file-is-compiled-for-system.
Calculate cross-correlograms with a wide range of options.

  ccg = ch_xcorr(hc_L,hc_R,fs)
  ccg = ch_xcorr(hc_L,hc_R,fs,'parameter',value)
  [ccg,ic] = ch_xcorr(...)

  ccg = ch_xcorr(hc_L,hc_R,fs) cross-correlates the input
  2-D matrices hc_L and hc_R over 10ms frame with a
  maximum lag of 1ms. It is assumed that the number of
  frequency channels is min(size(hc_L)) and hence hc_L and
  hc_R can be in either orientation. The
  cross-correlograms consist of cross-correlations for
  every frame and frequency channel. ccg has dimensions
  [lag,frequency,frame]. The function calculates running
  cross-correlations for every sample and integrates these
  cross-correlations over each frame. The number of
  frames frame_count is calculated thus:

  frame_count = ...
      floor((max(size(hc_L))-maxlag-1)/frame_length);

  The underlying cross-correlation algorithm is based on
  that proposed by Faller & Merimaa [1]. In this
  implmentation, the time constant of the backward
  infinite exponential window is given by tau (in
  samples).
  
  ccg = ch_xcorr(hc_L,hc_R,fs,'parameter',value) allows a
  number of options to be specified. The options are:

  ({} indicates the default value)

  'frame_length' : {round(0.01*fs)} | scalar
      The length of frames used to calculate for
      integrating cross-correlations.
  'noverlap' : {1} | scalar
      The number of frames over which to integrate the
      cross-correlations. Note that the frame count is
      reduced accordingly.
  'maxlag' : {round(0.001*fs)} | scalar
      The maximum lag of the cross-correlation.
  'tau' : {round(0.01*fs)} | scalar
      The time constant of the exponential window used to
      calculate running cross-correlations.
  'inhib' : {[]} | array
      Specificies an array with which to multiply the
      cross-correlations before they are integrated. The
      value defaults to an empty array, meaning that no
      inhibition will be applied.
  'ic_t' : {0} | scalar
      Specifies the interaural coherence (IC) threshold.
      Only samples for which the IC exceeds this threshold
      will be used to integrate cross-correlations. The
      algorithm calculates Interaural Coherence (IC)
      according to [1]. The value should be in the range
      [0,1];
  'norm_flag' : {0} | scalar
      Specifies whether the cross-correlograms are
      calculated using normalised cross-correlations. A
      non-zero value indicates that normalised
      cross-correlations are used.
  'inhib_mode' : {'subtract'} | 'multiply'
      Specify how the inhibition is applied. The default
      'subtract' will subtract inhib from the running
      cross-correlations; 'multiply' will multiply inhib
      with the running cross-correlations.

  [ccg,ic] = ch_xcorr(...) returns the calculated IC to
  the matrix IC. Although the matrix returned is the same
  size as hc_L, IC is only calculated for samples
  1:frame_count*frame_length, other values will be set to
  0.

  Algorithm

  See the enclosed documentation for more details on the
  workings of the algorithm and an important caveat.

  References

  [1] C. Faller and J. Merimaa, "Source localization in
  complex listening situations: Selection of binaural cues
  based on interaural coherence", The Journal of the
  Acoustical Society of America, vol. 116, pp.3075-3089,
  Nov. 2004.

  Further Reading
  
  C. Hummersone, R. Mason, and T. Brookes, "A comparison
  of computational precedence models for source separation
  in reverberant environments", in 128th Audio Engineering
  Society Convention, London, May 2010, paper 7981.

Acknowledgements

Check Whether Mex File Is Compiled For System inspired this file.

MATLAB release MATLAB 8.3 (R2014a)
MATLAB Search Path
/
/ch_xcorr
Other requirements Requires a C compiler and the following function: http://www.mathworks.co.uk/matlabcentral/fileexchange/27140-check-whether-mex-file-is-compiled-for-system.
Tags for This File   Please login to tag files.
Please login to add a comment or rating.
Comments and Ratings (1)
24 Oct 2010 Christopher Hummersone

whoops, you'll need this file: http://www.mathworks.com/matlabcentral/fileexchange/27140-check-whether-mex-file-is-compiled-for-system

Comment only
Updates
27 Jul 2010 1.1

Updated the mex function

06 Aug 2010 1.2

Fixed an issue with an un-specified max function. Thanks to Lars Bramsløw for pointing it out.

06 Aug 2010 1.3

Whoops, fixed another silly mistake, thanks to Lars Bramsløw

21 Oct 2010 1.4

Added additional 'NOVERLAP' parameter for integrating over several frames. Added note to documentation on limitations for normalised cross-correlation.

22 Oct 2010 1.5

Created additional documentation explaining the algorithm.

26 Oct 2010 1.7

Fixed a bug that gave negative numbers when using the 'subtract' option. Updated the documentation accordingly.

11 Nov 2010 1.8

removed need to framecount parameter input, now calculated automatically.

18 Apr 2011 1.9

Updated check for inhib matrix size and shape

16 Jun 2011 1.10

Updated documentation, added verbose comments through source code

04 Aug 2011 1.11

Added updated check_mex_compiled.m

17 Aug 2011 1.12

Fixed a bug whereby output IC would not be the same size as the input data. Added check to ensure that they are the same size and transposition. Updated help to include dims of output CCG.

18 Jan 2012 1.13

Some bug fixes.

26 Jun 2013 1.14

Output IC now reflects specified threshold.

20 Feb 2015 1.15

Made many simplifications to the C source code, and added error checks to m code.

24 Feb 2015 1.16

A few simplifications made to the C source code. Overhauled the interface: dramatically reduced the number of required arguments, and added other arguments as optional parameter/value pairs.

Contact us