Automatic determination of acoustic model topology using variational Bayesian estimation and clustering

• Published in: 2004 IEEE International Conference on Acoustics, Speech and Signal Processing Vol. 1; pp. I - 813
• Main Authors: Watanabe, T., Sako, L., Nakamura, F.
• Format: Conference Proceeding
• Language: English; Japanese
• Published: Piscataway, N.J IEEE 28.09.2004
• Subjects: Applied sciences; Bayesian methods; Clustering algorithms; Decision trees; Distributed computing; Exact sciences and technology; Hidden Markov models; Information, signal and communications theory; Iterative methods; Laboratories; Maximum likelihood estimation; Signal and communications theory; Signal processing; Signal representation. Spectral analysis; Signal, noise; Speech processing; Speech recognition; Telecommunications and information theory; Topology; Bayes estimation; Performance evaluation; Automatic classification; Mixture theory; Iterative method; Decision tree; Signal classification; Gaussian process; Covariance; Speech recognition; Phonetics; Speech processing; Posterior distribution
• ISBN: 9780780384842; 0780384849
• ISSN: 1520-6149
• DOI: 10.1109/ICASSP.2004.1326110

We describe the automatic determination of an acoustic model for speech recognition, which is very complicated and includes latent variables, using VBEC: variational Bayesian estimation and clustering for speech recognition. We propose an efficient Gaussian mixture model (GMM) based phonetic decision tree construction within the VBEC framework. The proposed method features a novel approach to reduce the unrealistically large number of computations needed for iterative calculations in the GMM-based decision tree method to a practical level by assuming that each Gaussian per state has the same occupancy and is represented by the same posterior distribution for the covariance parameter. The experimental results confirmed that VBEC automatically provided an optimum model topology with the highest performance level.