Design And Implementation Of Chinese Continuous Speech Recognition System Based On HTK

The purpose of speech recognition is to make the computer understand what people say. The theory of speech recognition has been well developed in the past 50 years. A lot of speech recognition algorithms and models have been already proved to be much effective by amounts of experimental results and practical operations. Speech recognition technology has been applied in various fields. This thesis explores the application of speech recognition theory used in Chinese continuous speech recognition.The basic process of speech recognition and the theoretical foundation of speech signal processing were firstly introduced. Moreover, the method and principles of voice activity detection and feature extraction were emphatically discussed. Then, as the main content of this thesis, Chinese continuous speech recognition was elaborated in depth in two aspects:In the way of pattern recognition, pronunciation characteristics of Chinese speech were concerned, parameters of Chinese speech recognition were extracted, and the corresponding speech recognition model was trained, while the experimental platform of Chinese continuous speech recognition was built. A continuous speech recognition experimental platform was built based on HTK (HMM Toolkit) using HMM (Hidden Markov Model) theory and a series of technologies such as MFCC (Mel-Frequency Cepstral Coefficients), mono-phone model, tri-phone model and Viterbi algorithm. Experiential results showed that as the HMM changed from mono-phone model to tri-phone mode, the recognition accuracy of statement-level and word-level were both increased, and reached a higher level after Tied-State Tri-phone model was employed. The recognition accuracy of statement-level increased from 76.00% to 96.00%,and the recognition accuracy of word-level increased from 90.67% to 98.00%.Another aspect of the thesis focused on software development, a Chinese continuous speech recognition simulation system used for ticketing was built. Firstly, the basic principle and composition of ATK (An Application Toolkit for HTK) was introduced. Then, the Chinese speech ticketing system was built with VS.NET as the platform and ATK as the development tool. At last, the corresponding testing experiment was finished. Experiential results revealed that it realized the basic function of a primary Chinese speech ticketing system.