Construction of model-space constraints
Tóm tắt
HMM systems exhibit a large amount of redundancy. To this end, a technique called eigenvoices was found to be very effective for speaker adaptation. The correlation between HMM parameters is exploited via a linear constraint called eigenspace. This constraint is obtained through a PCA of the training speakers. We show how PCA can be linked to the maximum-likelihood criterion. Then, we extend the method to LDA transformations and piecewise linear constraints. On the Wall Street Journal (WSJ) dictation task, we obtain 1.7% WER improvement (15% relative) when using self-adaptation.
Từ khóa
#Maximum likelihood linear regression #Covariance matrix #Maximum likelihood estimation #Hidden Markov models #Principal component analysis #Speech #Linear discriminant analysis #Piecewise linear techniques #Gaussian processes #VocabularyTài liệu tham khảo
10.1109/89.876308
bacchian, 2000, Using maximum likelihood linear regression for segment clustering and speaker identification, Proc of ICSLP, 4, 536
nguyen, 1999, Maximum likelihood Eigenspace and MLLR for speech recognition in noisy environments, Proc of Eurospeech, 6, 2519
10.1109/ICASSP.2001.940838
nguyen, 2000, EWAVES: an efficient decoding algorithm for lexical tree based speech recognition, Proc of ICSLP, 4, 286
duda, 1973, Pattern Classification and Scene Analysis
10.1006/csla.1995.0010
10.1109/89.848223