Based On Improving The Research And Design Of Mfcc Speech Recognition System

As the development of the information science and embedded technology, the research of speech recognition technology reach a new level. Using modern methods to study the speech recognition technology, it can effectively help us produce, storage and retrieval of speech signal, and it also promotes social development with great significance.This paper put forward a speech recognition system based on improved Mel Frequency cepstrum parameters(MFCC) and realize its embedded hardware design. The speech recognition algorithm is introduced and realized in details. Before specific speech recognition, this paper introduces some short-term operations of the speech recognition, including the short-term average energy, the short-term average zero rates, the short-term autocorrelation and short-time average magtitude difference function. These operations will make a good bedding for the followed speech recognition algorithm.The endpoint detection which is difficult in the speech recognition, considering the situation of the direct current interference and long end tone mistaken judgment,we put forward a improved double threshold method.for the pitch detection issue in speech recognition,we put forward a improved circulation Aeverage Magnitude Difference Function(AMDF) method. In order to obtain the best speech characteristic parameters, we put forward the improved MFCC method which is based on the traditional MFCC and its medium to high frequency band frequency characteristics is improved. Experimental results prove that the improved characteristic parameters extraction method increases the speech recognition rate. In the choice of the speech recognition model, we introduce the Dynamic Time Warping Model (DTW),the Hidden Markov Model (HMM) and Artificial Neural Networks Model (ANN).Using the DTW model for speech recognition,the model algorithm is simple and the recognition rate is high.In this paper, the hardware design is also introduced in details. With samsung's S3C2440 processor as the core, the processor has strong hardware resources. This paper gives the connection circuit,including the processor with SDRAM, FLASH, Serial ports, JTAG, Power supply, external AD, LCD, the sound part and so on.In this paper, the specific recognition algorithms simulation is conducted in MATLAB software, and achieved good simulation results. Results show us that the algorithm can be widely used in speech recognition.To some extent,it is stimulative to the speech recognition technology.