Research And Improvement Of Chinese Speech Recognition And Tone Extraction

Speech Recognition is a very important field in Pattern Recognition subject, which can help people communicate with computer. Since Chinese has the most population in the world, Chinese Speech Recognition has very good future in application. In this paper, after analyzing the system flow of Chinese Isolated Word Recognition, we focus on Speech Recognition algorithm and Speech Pitch Extraction parts.Firstly, we do research on entire system of Speech Recognition, especially Signal Pre-processing part. It is the fundamental of a good Speech Recognition system. The effect of Endpoint Detection can influence the results of the whole system, so we pay attention to the requirements of the Signal pre-processing module and make a solution, which has already been used in a real project.Secondly, we do research on DTW and HMM algorithms, which are widely used in Speech Recognition field. We deeply talk about the principle and applications of these algorithms, and establish real systems using them respectively. The results verify the performance of them on Chinese Isolated Word Recognition and Small Vocabulary Recognition, which can be a reference for others.Finally, we do research on Pitch Extraction, and improve the method for Chinese syllable. By analyzing the structure of Chinese syllable and features of phonemes, we advance the Tone Extraction by using Phonemes Separation. According to the results of experiments on different types of syllable, the improved system has better performance than original system. At last, we define the applying scale of the improved algorithm.