Audio Demonstrations

Publication:

Kim, G., Lu, Y., Hu, Y. and Loizou, P. (2009). "An algorithm that improves speech intelligibility in noise for normal-hearing listeners," Journal of Acoustical Society of America, 126(3), 1486-1494. [pdf]

Code:

MATLAB code for extraction of amplitude modulation spectrogram (AMS) features [AMS_features.rar].

GMM training was done using the GMMBAYES MATLAB toolbox available from here.

Example Sentence 1 (from IEEE corpus)

Model	Babble	Factory noise	Speech-shaped noise
Unprocessed (-5 dB SNR)	bab_m5db_x1.wav	fac_m5db_x1.wav	sps_m5db_x1.wav
Processed via sGMM (single noise trained GMMs)	bab_m5db_x1_sGMM.wav	fac_m5db_x1_sGMM.wav	sps_m5db_x1_sGMM.wav
Processed via mGMM (multiple noise trained GMMs)	bab_m5db_x1_mGMM.wav	fac_m5db_x1_mGMM.wav	sps_m5db_x1_mGMM.wav

Sentence 1 (in quiet)

Example Sentence 2

Model	Babble	Factory noise	Speech-shaped noise
Unprocessed (-5 dB SNR)	bab_m5db_x2.wav	fac_m5db_x2.wav	sps_m5db_x2.wav
Processed via sGMM	bab_m5db_x2_sGMM.wav	fac_m5db_x2_sGMM.wav	sps_m5db_x2_sGMM.wav
Processed via mGMM	bab_m5db_x2_mGMM.wav	fac_m5db_x2_mGMM.wav	sps_m5db_x2_mGMM.wav

Sentence 2 (in quiet)

Example Sentence 3

Model	Babble	Factory noise	Speech-shaped noise
Unprocessed (-5 dB SNR)	bab_m5db_x3.wav	fac_m5db_x3.wav	sps_m5db_x3.wav
Processed via sGMM	bab_m5db_x3_sGMM.wav	fac_m5db_x3_sGMM.wav	sps_m5db_x3_sGMM.wav
Processed via mGMM	bab_m5db_x3_mGMM.wav	fac_m5db_x3_mGMM.wav	sps_m5db_x3_mGMM.wav

Sentence 3 (in quiet)

Speech Processing Lab | Cochlear Implant Lab