Chinese Name Speech Classification Using Fisher Score Based on Continuous Density Hidden Markov Models

Over the last years significant effort has been made to improve the performance of speech recognition. The Fisher Kernel has been suggested as good ways to combine and underlying generative model in the feature space and discriminant classifiers such as SVMs. Chinese name speech patterns are difficult to be classified especially when they are similar in pronunciation. Continuous density hidden Markov model(CHMM) is state-of-the-art method to process this difficulty. A procedure was proposed in this paper to derive the Fisher score from CHMM, and compare it with traditional generative models and Gaussian mixture model(GMM) based Fisher score in Chinese name speech recognition. The result shows that CHMM based Fisher score classified by SVMs receives the best performance.