Orthogonal Subspace Combination Based on the Joint Factor Analysis for Text-Independent Speaker Recognition
To apply a joint factor analysis (JFA) in a multiple channel circumstance, this paper proposes an orthogonal subspace combination method for a text-independent speaker recognition system. On the condition of multiple channels, the subspace loading matrix estimated by a mixed data corpus suffers from the data masking effects. And the subspace loading matrix estimated by a simple combination method has a drawback of subspace overlapping. To overcome these problems, this paper presents an orthogonal subspace combination method. The proposed method is based on a proper approximation of the core computation of the JFA and makes use of the Gram-Schmidt orthogonalization. On the NIST SRE 2008 core tasks corpus, the proposed method has a better performance.
KeywordsGaussian mixture models joint factor analysis subspace combination text-independent speaker recognition
Unable to display preview. Download preview PDF.
- 3.National Institute of Standards and Technology. NIST speaker recognition evaluation, http://www.itl.nist.gov/iad/mig/tests/spk/2008/index.html
- 5.Vogt, R., Sridharan, S.: Experiments in Session Variability Modelling for Speaker Verification Acoustics. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, pp. 897–900. IEEE Press, New York (2006)Google Scholar