The Research Of Speech-Recognition And The Basic Realization

The speech-recognition is letting the computer understand our human being and react rightly. The speech-recognition technology is one of the most important directions of computer technology. Since 90's the speech-recognition systems has been developed as an integrated theory, on the other hand the speech-recognition systems run on PC have been developed so well and have gotten some success. Although the research of speech-recognition has been arrived the state of commodity, the basic theory is quite perfect and lots of products come forth in succession, but in many special field, because of the circumstance and the industry, we often need develop the system specially.Firstly, this paper introduces the development of speech recognition, elucidates the background and significance of the research and the difficulties we faced when we want to make it use broadly.Secondly, according to the model of speech-recognition system it describes the fundamental of every part (including pretreatment, end point detection, template production and template matching) such as the speech data signal process, the mode recognition and discusses some method of the realization of speech-recognition on PC.Thirdly, it discusses some different methods of realizing the speech-recognition system and describes how to use Visual C++6.0 to implement a speaker independent, small vocabulary, isolated-speech-recognition system which uses training and recognition methods of DTW model and run on PC (the speech-recognition control interface of a simulated line-query system), it consistent with the model of the speech-recognition system. In the phase of training, it gets the sampling data from the wave files which were stored in the voice library by using the MCI functions. Then calculates the character vector (12 ranks of LPC and LPCC) and trains them by clustering method, so we get the templates used by speech-recognition, this templates were stored in the template library. In the state of recognition, after calculating the character vector of input voice, we compare it with the character vectors of templates, and then find the best one or refuse it. On the basis of the above we put forward some improved schemes such as the linear prediction in lower SNR (Signal Noise Ratio) environment, the endpoint detection method and the threshold decision by ANN according to the real circumstance.At last the paper indicates the directions of improvement. Through the test and research of the actual speech recognition system, it carries out the fundamental and exploring research for the further application of speech recognition system.