Research On Speech Recognition System Based On The Improved Hybrid HMM/SVM Framework

Speech recognition is an automatic recognition process of a speaker that based on the voice parameter in the speech waveform, which reflects the speaker's physiological and behavioral characteristics. It is an important research orientation of speech signal processing. As a biological recognition technology, it has a wide range of application and has been studied more and more.This dissertation presents the theories and methods of Hidden Markov Models (HMM) and Support Vector Machine (SVM) adopted in speech recognition technology. The originality and main work on research of this dissertation are listed as follows:(1) The basic theory of speech recognition technology, continuous recognition technology and its state-of-the-art are investigated.(2) The theory of HMM and its main algorithm are studied deeply. On this basis, the author has improved the training methods and proposed a modified model for HMM-based speech recognition system framework.(3) The theoretical basis of statistical learning theory of SVM and the application of SVM in pattern recognition are researched.(4) After a series of study, the author has proposed a hybrid HMM/SVM model and applied it to digital speech recognition system.