论文信息 - Application of the MLVQ1 in speaker identification

Application of the MLVQ1 in speaker identification

In this paper we describe a new method, in automatic speaker recognition, based on modified LVQ1 (MLVQ1) and using 3 prosodic features: the mean of the pitch, the original duration and the low-frequency energy. For this purpose, we conceived a new metric, optimized in automatic speaker recognition, which we called ODHEF. The tests of speaker recognition are done in Arabic corpus with 2 different sets: a closed set and an open set. The results show that the prosodic features are relevant and that the modified LVQ1 (MLVQ1) method is interesting in text-dependent speaker identification.

Halim Sayoud | Siham Ouamour-Sayoud | Malika Boudraa

[1] Ivan Magrin-Chagnolleau,et al. Second-order statistical measures for text-independent speaker identification , 1995, Speech Commun..

[2] Y. Ishida,et al. Speaker-independent spoken digits recognition using LVQ , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).

[3] J. Wolf. Efficient Acoustic Parameters for Speaker Recognition , 1972 .

[4] Younès Bennani,et al. Neural networks for discrimination and modelization of speakers , 1995, Speech Commun..

[5] B. Atal. Effectiveness of linear prediction characteristics of the speech wave for automatic speaker identification and verification. , 1974, The Journal of the Acoustical Society of America.

[6] B. Atal. Automatic Speaker Recognition Based on Pitch Contours , 1969 .

[7] Douglas A. Reynolds,et al. Experimental evaluation of features for robust speaker identification , 1994, IEEE Trans. Speech Audio Process..

[8] Jorma Laaksonen,et al. LVQ_PAK: The Learning Vector Quantization Program Package , 1996 .