DocumentCode :
2696634
Title :
Improvements in MLLR-Transform-based Speaker Recognition
Author :
Stolcke, Andreas ; Ferrer, Luciana ; Kajarekar, Sachin
Author_Institution :
Speech Technol. & Res. Lab., SRI Int., Menlo Park, CA
fYear :
2006
fDate :
28-30 June 2006
Firstpage :
1
Lastpage :
6
Abstract :
We previously proposed the use of MLLR transforms derived from a speech recognition system as speaker features in a speaker verification system. In this paper we report recent improvements to this approach. First, we noticed a fundamental problem in our previous implementation that stemmed from a mismatch between male and female recognition models, and the model transforms they produce. Although it affects only a small percentage of verification trials (those in which the gender detector commits errors), this mismatch has a large effect on average system performance. We solve this problem by consistently using only one recognition model (either male or female) in computing speaker adaptation transforms regardless of estimated speaker gender. A further accuracy boost is obtained by combining feature vectors derived from male and female vectors into one larger feature vector. Using 1-conversation-side training, the final system has about 27% lower decision cost than a state-of-the-art ccpstral GMM speaker system, and 53% lower decision cost when trained on 8 conversation sides per speaker
Keywords :
Gaussian processes; maximum likelihood estimation; regression analysis; speaker recognition; training; transforms; 1-conversation-side training; MLLR transform; average system performance; speaker verification system; speech recognition system; state-of-art cepstral GMM speaker system; Cepstral analysis; Costs; Lattices; Maximum likelihood decoding; Maximum likelihood linear regression; Mel frequency cepstral coefficient; NIST; Predictive models; Speaker recognition; Speech recognition;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Speaker and Language Recognition Workshop, 2006. IEEE Odyssey 2006: The
Conference_Location :
San Juan
Print_ISBN :
1-424400471-1
Electronic_ISBN :
1-4244-0472-X
Type :
conf
DOI :
10.1109/ODYSSEY.2006.248089
Filename :
4013506
Link To Document :
بازگشت