Title :
A modified group vector quantization algorithm for speaker identification
Author :
Abu El-Yazeed, M.F. ; Abdel Kader, Nemat S. ; El-Henawy, M.M.
Author_Institution :
Dept. of Electron. & Comm., Cairo Univ., Giza
Abstract :
In this paper, a modification to the group vector quantization (GVQ) discriminative training algorithm is proposed to train VQ codebooks for closed set speaker identification. The proposed algorithm, referred to as modified GVQ (MGVQ), shifts the decision surfaces between speakers smoothly toward the Bayes limits. This is achieved by varying the learning rate during training iterations. The proposed MGVQ algorithm achieves higher speaker identification rate compared to the standard GVQ
Keywords :
speaker recognition; speech coding; vector quantisation; Bayes limit; VQ codebooks; closed set speaker identification; decision surfaces; discriminative training algorithm; group vector quantization; learning rate; modified GVQ; text independent speaker identification; training iteration; Cepstral analysis; Clustering algorithms; Feature extraction; Hidden Markov models; Iterative algorithms; Linear predictive coding; Spatial databases; Speech; Testing; Vector quantization; Group vector quantization; speaker identification; text independent;
Conference_Titel :
Circuits and Systems, 2003 IEEE 46th Midwest Symposium on
Print_ISBN :
0-7803-8294-3
DOI :
10.1109/MWSCAS.2003.1562365
