Noise Environment Of Chinese Continuous Speech Recognition Technology Research,

Speech recognition has been researched for more than half a century, and great progress has been made. Though current speech recognition system has been achieved very high recognition accuracy in the clean speech environment, the ubiquitous noise significantly reduces the performance of the system. Therefore, Anti-noisy technology is a very crucial problem for speech recognition in the application.The main study of this paper is the key technologies of Chinese continuous speech recognition. Firstly the paper introduces the principle of speech recognition, the composition and key technologies of speech recognition system, and so on. And then it introduces the classification of noise and various anti-noisy technologies. On the basis, the main work of this paper is as follows:1) A medium-vocabulary and speaker-independent Chinese continuous speech recognition system is achieved on a personal computer. This system chooses the syllable as recognition unit, the MFCC as feature parameters, and the Hidden Markov Model as recognition model. And then we make experiments on the system to analyze the performance of the whole system.2) How to accurately detect the start and the end point of syllable is a very important step in Chinese continuous speech recognition. The existing Chinese continuous speech endpoint detection method can obtain very high accuracy in clean speech environment, while the accuracy will be significantly reduced in noisy environment. According to the feature of Chinese continuous speech and noise, an improved method based on vowel is proposed in this paper, which can effectively improve the endpoint detection accuracy in noisy environment.3) What the speech recognition system processes are feature parameters whose anti-noisy performance can help improve the system's performance. Based on analyzing the traditional Mel frequency cepstral coefficients extraction, wavelet packet and weighted filter, a new method of feature extraction is proposed. The experimental results show that the improved feature parameters can achieve higher recognition rate and better anti-noisy performance than the traditional feature parameters.