Wavelet Transform to Hybrid Support Vector Machine and Hidden Markov Model for Speech Recognition
Yu Shao, Chip-Hong Chang · 2005
Recent work in machine learning has focused on models, such as the support vector machine (SVM), that automatically control generalization and parameterization as part of the overall optimization process. In this paper we construct a wavelet transform to hybrid support vector machine and hidden Markov models (WSVM/HMM) speech recognition system to deal with arbitrary nonlinear functions. The proposed hybrid system has overcome the problem of acoustic modeling in state-of-the-art speech recognition systems that usually relies on HMM with Gaussian emission densities. HMM suffer from intrinsic limitations, mainly due to their arbitrary parametric assumption. Artificial neural networks appear to be a promising alternative, but they have historically failed as a general solution to the acoustic modeling problem. The proposed system has successfully exploited the benefits of the wavelet technique, HMM and SVM in a unified framework. In this hybrid system, the WSVM is trained to estimate the emission probabilities of the states of the HMM. Simulations on benchmark databases show that speech recognition systems built around the hybrid WSVM/HMM provide excellent word recognition ratio and have performance superior to many other systems.