Research On Speech Recognition System Base On Dynamically Adapting Environment Endpoint Detection

Speech signal is the essential medium of information transmission and emotion intercourse, and apperceive of human acoustical apparatus to sound and liberation. It is also the most important, efficient, common, and convenient tools for human communications. Speech Signal Processing is one of the booming information science subjects, which refer to many fields of science. Endpoint Detection is one of the most important part in a speech recognition system. Efficient Endpoint Detection is able to increase system processing speed, enhance system real-time concurrence, avoid disturbance from noise and mute section, and enhance the recognition performance. Nowadays, most of the endpoint detection algorithms can not perform well under a low SNR.Endpoint Detection is a key technique in Speech Recognition. Nowadays most kinds of Endpoint Detection algorithm perform well in the environments of high Signal-to-Noise(SNR), but they all perform worse while SNR is reducing which could not adapt the complicated out-door environment with mess of noise. For the random city of White Gaussian Noise in outdoor environments, it is advanced to use GMM to describe the White Gaussian Noise in our common environments. By the investigation of GMM and noises, it is advanced that the Endpoint Detection algorithm basing on GMM, and the flow of the algorithm is expatiated in detail. And the algorithms of each module in a Speech Recognition are investigated and improved, including Speech Enhancing, Feature Extraction and Template Matching. Then it designed all of the modules and designed a integrated Speech Recognition system which would automatically adapt the change of environments.The system is achieved to the recognition of single word in complicated environment, and update by itself according to the environment change. However, its study speed is limited for the amount of calculation, which needs to be improved.