Skip to content

iiscleap/FeatureExtractionUsingFDLP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

%*****************************************************************
% Sriram Ganapathy
% Center of Language and Speech Processing 
% Johns Hopkins University
% ganapathy@jhu.edu
%*****************************************************************
% 11-Jan-2012
% See the file COPYING for the licence associated with this software.
******************************************************************
Ref:
S. Ganapathy, S. Thomas and H. Hermansky, "Temporal envelope compensation for robust phoneme recognition using modulation spectrum ", Journal of Acoustical Society of America, Dec. 2010.
S. Thomas, S. Ganapathy and H. Hermansky, "Recognition Of Reverberant Speech Using Frequency Domain Linear Prediction", IEEE Signal Processing Letters, Dec 2008.
S. Ganapathy, S. Thomas and H. Hermansky, "Feature Extraction Using 2-D Autoregressive Models For Speaker Recognition", IEEE Speaker Odyssey, Jun. 2012.
******************************************************************
DESCRIPTION

The FOLDER CONTAINS MATLAB FUNCTIONS FOR FEATURE EXTRACTION USING FREQUENCY DOMAIN LINEAR PREDICTION (FDLP). THE VARIOUS EXTRACTION SCHEMES ARE
1. FDLP BASED SPECTRAL FEATURE EXTRACTION (FDLP-S) - SHORT-TERM FEATURES SIMILAR TO MFCC
2. FDLP BASED MODULATION FEATURE EXTRACTION (FDLP-M) - LONG-TERM MODULATION FEATURES
3. FDLP + PLP FOR 2-D AUTO-REGRESSIVE MODELLING (FDLP_PLP2) - SHORT-TERM FEATURES SIMILAR TO PLP 

FOR USING ALL THE FUNCTIONALITIES IN FDLP-M, PLEASE DOWNLOAD AND COMPILE ADAPTIVE COMPRESSION  LOOP MEX IMPLEMENTATION FROM http://medi.uni-oldenburg.de/download/demo/adaption-loops/adapt_loop.zip 

THE ADAPT_M MEX FILE MUST BE PLACED IN THE CURRENT DIRECTORY.
*******************************************************************
USAGE

IN MATLAB: 
FEAT = FDLP_FEAT(SAMPLES,CONFIG_FILE)

THE DESCRIPTION OF THE CONFIG FILE ENTRIES IS PROVIDED AT THE BOTTOM OF THIS README. 
*******************************************************************
EXAMPLE

IN MATLAB
fid = fopen ('fcjf0_sx37.raw', 'rb','l');
[x,cnt] = fread(fid,inf,'int16');
fclose (fid);

feats = fdlp_feat(x,'matlab.config');
*******************************************************************
ACKNOWLEDGEMENTS

SOME OF THE FUNCTIONS IN THIS IMPLEMENTATION MODIFY THE ORIGINAL IMPLMENTATION OF PLP FEATURES FROM DAN ELLIS.
http://labrosa.ee.columbia.edu/matlab/rastamat/
THE INITIAL IMPLEMENTATION OF THE FDLP TECHNIQUE CAME FROM MARIOS ATHINEOS
http://www.ee.columbia.edu/~marios/
******************************************************************
COPYRIGHT

Read the file COPYING for copyrights.
*******************************************************************

Decription of the configuration parameters


% Signal Parameters
SAMPLE_RATE             % Sampling Rate
FDLP_LEN                % FDLP frame length in seconds (1 second)
FLAG_FULL_SIG           % Using the entire signal in one frame length (useful for short speech files of 2-3s)

% Type of Feature
FEAT_TYPE               % Type of features (fdlps=1/fdlpm=2/fdlp_plp2=3)
AXIS	                % Frequency Axis   (bark=1/mel=2/linear=3)
MEL_WARP 	        % Applies only for converting linear to mel-scale  
DCT_LOW	                % Begin of Frequency Axis in DCT domain
DCT_HIGH	        % End of Frequency Axis in DCT domain
SKIP_BANDS 	        % Skip the initial bands in feature computation

% Feature configuration
FLAG_GAIN_NORM          % Gain normalization 
MODEL_ORDER             % Model order per sub-band per second
FLAG_LP_TYPE            % Type of linear prediction (autocorr=1/ls=2)
FLAG_WIENER             % Apply Wiener filtering in feature

% Modulation feature configuration
TEMP_CEPS 	        % Number of modulation components per band  
FLAG_ADAPT              % Including static and adaptive compression
                        % Requires mex file adapt_m

% Spectral feature parameters
SPEC_CEPS	        % Number of cepstral components
FLAG_C0                 % Flag set to 1 for using C1-C13
SPEC_FR_LEN             % Frame length for spectral frame (ms)
SPEC_FR_SHIFT           % Frame shift for spectral fram (ms)   
FLAG_DELTA  	        % Apply Delta and Acceleration 
*******************************************************

About

Feature Extraction using FDLP

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages