The Study Of Car Speech Recognition Based On The Specific Human And Small Vocabulary

Speech recognition and processing in human-machine interfaces technology is widespread concerned. It's application makes people's lives more convenient. People can operate the device only by the command of the voice. The device which is a voice-recognition system on the Internet, communications, military, national defense and etc is of very important value. Similarly, the speech recognition technology in the platform of vehicle, it must make the driving is more flexible, more security and comfortable.This paper is a paper based on the DTW model of speech recognition technology and it introduces the basic methods of speech recognition. That the speech recognition is improved and optimized is to be applied to vehicle simulation systems. This paper is focusing on the optimization-based endpoint detection, combining with variable window length, two-voice activity detection threshold and then making optimal path selection, taking a relaxing way to the begin and the end, so speech recognition will be more accurate. The simulation and experiment can be seen that these methods could improve the accuracy of speech recognition. And tentative on the fuzzy algorithm is applied to speech recognition model. Specifically, first, the processing of DTW model based on speech recognition including how to remove the noise, the speech feature parameter's extraction and the inter-transform of the signal between frequency domain and time domain and the basic theory of speech recognition, this paper makes the introduction and analysis to that. While determining the isolated word speech recognition applications and the DTW model, it has been improved and optimized the recognition algorithm and it realizes the system. The simulation will be improved compared the optimal algorithm speech recognition with the previous traditional method. This paper analyzed how the voice recognition algorithm improved. When at the frame processing, quiet segments, voice segments and transient segments get the different windows to process. In the quiet section of the signals of voice, we can use longer window length came to pick up the frame. In the transition section of the voice, we can use shorter window length to pick up the frame. In the voice section, we can use regular window length to pick up the frame. At the same time we also could use a double threshold method for endpoint detection, which combines short-time average energy and the short-term zero crossing rate, we take the low threshold and high threshold to limit the value of the starting point and end point. In the specific process of DTW algorithm, using dynamic warping and relaxation methods in the endpoint, selecting the optimal path, so it can get a more accurate voice match. In specific applications, we combine the system with the highest minimum cost of the program to achieve it. In the entire application of the algorithm, this paper completes the implementation. Speech recognition process is completely self-made program, in the control of the car, the method of sending wave band division used to deal with the different car actions. This paper proposes the point which is needed to improve in speech recognition and presents the future prospect of speech recognition.