MATLAB code for the SNRloss and SNRLESC measures described in “SNR loss: A
new objective measure for predicting speech intelligibility of noise-suppressed speech”
Speech Comm.
Philipos C. Loizou and Jianfen Ma
SYSTEM REQUIREMENTS
1. MATLAB v. 6x or higher
2. MATLAB’s Signal Processing Toolbox
FILE USAGE
Usage:
loss = SNRloss(cleanFile.wav,enhancedFile.wav, mtype);
mtype – speech material type: 1-consonant weights, 2-sentence weights
cleanFile.wav - clean speech file name
enhancedFile.wav - enhanced speech file name
loss – the objective intelligibility score of SNRloss measure
[SNRLESC_h,SNRLESC_m,SNRLESC_l] =
SNRLESC(cleanFile, enhancedFile,noiseFile,mtype)
mtype – speech material type, 1-consonant weights, 2-sentence weights
cleanFile.wav - clean speech file name
enhancedFile.wav - enhanced speech file name
noiseFile.wav - masking noise speech file name
SNRLESC_h – the objective intelligibility score of SNRLESCHigh measure
SNRLESC_m – the objective intelligibility score of SNRLESCMid measure
SNRLESC_l – the objective intelligibility score of SNRLESCLow measure
EXAMPLE
A sample sentence (S_03_01.wav) is included from the IEEE database, along with the
processed (via subspace KLT algorithm) sentence (S_03_01_babble_sn0_klt.wav).
The noisy file was originally corrupted at 0 dB SNR (babble).
To call the above measures, type:
loss=SNRloss('S_03_01.wav','S_03_01_babble_sn0_klt.wav',2);
[SNRLESC_h,SNRLESC_m,SNRLESC_l]= SNRLESC('S_03_01.wav',
'S_03_01_babble_sn0_klt.wav','S_03_01_babble_sn0_klt_n.wav',2);
