Research And Implementation Of Chinese Isolated Words Recognition Based On ARM Platform

Speech recognition technology is to make the computer understand human language, that means it will be in accordance with the corresponding command execution and realize the humanized service for the purpose. With the rapid development of modern electronic information technology, voice recognition technology is used widely, such as car intelligent control and service, which can reflect this value. When the small car running, the goal of speech recognition system can tell the driver commands, which can identify the liberation of the driver not directly associated with driving operation. In order to concentrate, it can not only facilitate the driver, but also greatly improve the driving safety.This paper bases on the characteristic parameters and the selection of recognition model simulation research of Chinese isolated words speech signal. We use Samsung S3C2440microprocessor for isolated word speech recognition system, the Linux as a complete embedded speech recognition of embedded operating system, Qtopia2.2.0as a design system of human-computer interaction interface graphical system.At first,this paper summarizes the basic theory of speech signal processing, including speech recognition system overview, the front end of the speech signal processing module, the module and voice signal sampling and quantization, pretreatment, frame and window, speech signal endpoint detection, speech signal feature extraction module, including the extraction of characteristics and linear prediction cepstrum coefficient of LPCC, Mel frequence cepstrum coefficient of MFCC, Vector Quantization module, the module. Furthermore this paper introduces the principle of vector quantization distortion measure, and isolated words code book LBG algorithm is designed, Speech recognition module modeling method of reference template, including accidental template training method, the average template training methods before using BP model of the data processing module. Finally we introduce methods:speech recognition model including the most classic of isolated word speech recognition, which is the most basic Dynamic Time Wraping (DTW) algorithm and compare artificial neural network (BP) algorithm which is widely used in recent years.In view of the speech signal characteristic parameters of our system is analyzed through simulation experiments, and the choice of recognition model related content. We select the DTW recognition machine as pattern matching method of system identification, MEL frequency cepstrum coefficient MFCC as recognition parameters, on the basis of the demand of the embedded platform, aiming at the limit and the requirement of the real-time embedded system resources, in the form of text directly read identification reference template, improving the speed of recognition. Secondly, in order to make test results more intuitive, this paper chooses the Qtopia graphical system, which can be used in the QT platform of voice recognition application code and migration to the target board, and then sets up a speech recognition system based on ARM platform, a convenient operation, which establishs a cross-compilation environment, transplants the Uboot, Linux operating system transplant and the root file system. Finally, through the test of the whole system, the design of the system to achieve the goal of intelligence and human-computer interaction, the recognition performance of the system was tested.Embedded system as a platform for isolated words speech recognition is easy to use and suitable for application in different places. This thesis finally summarizes the thesis work and points out some of the limitations of the method.