Segmental LVQ3 training for phoneme-wise tied mixture density HMMS

This work presents training methods and recognition experiments for phoneme-wise tied mixture densities in hidden Markov models (HMM). The system trains speaker dependent, but vocabulary independent, phoneme models for the recognition of Finnish words. The Learning Vector Quantization (LVQ) methods are applied to increase the discrimination between the phoneme models. A segmental LVQ3 training is proposed to substitute the LVQ2 based corrective tuning as a parameter estimation method. The experiments indicate that the new method can provide the corresponding recognition accuracy, but with less training and more robustness over the initial models. Experiments to upscale the current system by introducing context vectors and larger mixture pools show up to 40 % reduction of recognition errors compared to the earlier results in [10].

[1]  Mikko Kurimo.  Corrective tuning by applying LVQ for continuous density and semi-continuous Markov models , 1994, Proceedings of ICSIPNN '94. International Conference on Speech, Image Processing and Neural Networks.

[2]  Chin-Hui Lee,et al.  Segmental GPD training of HMM based speech recognizer , 1992, [Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[3]  Shigeru Katagiri,et al.  GPD training of dynamic programming-based speech recognizers , 1992 .

[4]  Biing-Hwang Juang,et al.  New discriminative training algorithms based on the generalized probabilistic descent method , 1991, Neural Networks for Signal Processing Proceedings of the 1991 IEEE Workshop.

[5]  Teuvo Kohonen,et al.  Improved versions of learning vector quantization , 1990, 1990 IJCNN International Joint Conference on Neural Networks.

[6]  Mikko Kurimo,et al.  Hybrid training method for tied mixture density hidden Markov models using learning vector quantization and Viterbi estimation , 1994, Proceedings of IEEE Workshop on Neural Networks for Signal Processing.

[7]  Teuvo Kohonen,et al.  LVQ-based speech recognition with high-dimensional context vectors , 1992, ICSLP.

[8]  Mikko Kurimo,et al.  Combining LVQ with continuous-density hidden Markov models in speech recognition , 1992, Optics & Photonics.

[9]  Xuedong Huang,et al.  Semi-continuous hidden Markov models for speech signals , 1990 .

[10]  Kunio Nakajima,et al.  An optimal discriminative training method for continuous mixture density HMMs , 1990, ICSLP.

[11]  Teuvo Kohonen,et al.  The self-organizing map , 1990 .

[12]  Lawrence R. Rabiner,et al.  A segmental k-means training procedure for connected word recognition , 1986, AT&T Technical Journal.

[13]  E. McDermott,et al.  A hybrid speech recognition system using HMMs with an LVQ-trained codebook , 1990 .