Speaker Identification in Total Variability Space

Abstract:

Gaussian Mixture Model-Universal Background Model based approaches have been popular used for speaker identification task. But in real complex environment the identification system performs too much worse than in laboratory, and the main reason is the mismatch of the training and testing channel and also the variability of the speaker himself. In this paper we introduce i-vector to the speaker identification system. In i-vector approach, a low dimensional subspace called total variability space is used to estimate both speaker and channel variability. Baum-Welch statistics are first computed over the given UBM to estimate the total variability. From the experiment results, we obtain 2.44% relative accurate identification rate improvement when using total variability space to compensate the mismatch of the variabilities from both the speaker and channel.

• Periodical: Applied Mechanics and Materials (Volumes 401-403)
• Pages: 1489-1492
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.401-403.1489
• Online since: September 2013
• Authors: Qiang Li, Ming Bing Zhao, Yong Feng
• Keywords: GMM-UBM, MFCC, Speaker Identification, Total Variability

References

[1] D. A. Reynolds, T. F. Quatieri, R. B. Dunn, Speaker Verification Using Adapted Gaussian Mixture Models, Digital Signal Processing , 2000, 10(1), 19-41.

DOI: 10.1006/dspr.1999.0361

[2] W. M. Campbell, D. E. Sturim, D. A. Reynolds, SVM based speaker verification using a GMM supervector kernel and NAP variability compensation, in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process., Toulouse, France, 2006, vol. 1, p.97–100.

DOI: 10.1109/icassp.2006.1659966

[3] P. Kenny, G. Boulianne, P. Ouellet, Joint factor analysis versus eigenchannels in speaker recognition, IEEE Trans. Audio, Speech, Lang. Process. , vol. 15, no. 4, p.1435–1447, May (2007).

DOI: 10.1109/tasl.2006.881693

[4] N. Dehak, R. Dehak, P. Kenny, Support vector machines versus fast scoring in the low-dimensional total variability space for speaker verification, Interspeech, Brighton, UK, Sept (2009).

DOI: 10.21437/interspeech.2009-385

[5] A. Kanagasundaram, R. Vogt, D. Dean, i-vector Based Speaker Recognition on Short Utterances, Interspeech, Brighton, UK, (2011).

DOI: 10.21437/interspeech.2011-58

[6] D. A. Reynolds, A gaussian mixture modeling approach to text-independent speaker identification, Ph.D. thesis, Georgia Institute of Technology, August (1992).

[7] N. Dehak, P. Kenny, R. Dehak, Front-End Factor Analysis for Speaker Verification, IEEE Trans. on Audio, Speech and Language Processing, vol. 19, pp.788-798, May (2011).

DOI: 10.1109/tasl.2010.2064307