Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing > Speech Recognition and equipment

Research of Speech Recognition of Isolated Word Under Embedded Linux Operation

Author CaiHu
Tutor ZouXueCheng
School Huazhong University of Science and Technology
Course Software Engineering
Keywords Speech Recognition Dynamic Time Warping Hidden Markov Embedded Audio device driver Linux
CLC TN912.34
Type Master's thesis
Year 2009
Downloads 125
Quotes 2
Download Dissertation

Speech recognition and embedded hot today, its products very broad application prospects, speech recognition applications to embedded systems, it is able to play the advantages of both. This paper studies the embedded isolated word speech recognition, the main research design speech recognition algorithms: dynamic time warping algorithm DTW and hidden Markov algorithm HMM, then write based on Linux and ARM chip audio driver and tested it. In speech recognition algorithm, first create a model of the utterance of the speech recognition system, then voice for pretreatment, i.e., the digital signal preemphasis the stars frame processing window, and then extract characteristic parameters of the speech signal, a characteristic parameter extraction method linear prediction cepstral coefficients and beauty Seoul cepstral coefficients, and then next to the feature vectors extracted from the vector quantization, thus compressing the data. Finally, focus on analysis and design of dynamic time warping DTW (Dynamic Time Warping) recognition algorithm and hidden Markov HMM (Hidden Markov Model) recognition algorithm. The DTW mainly by finding the dynamic time warping function to make an error between the input module and the reference module to a minimum, thereby draw the similar error of the speech recognition. Hidden Markov HMM recognition algorithm is the solution to seek the probability of the observed values, find the best state sequence and under what circumstances observed maximum probability. The basis of the above two models, Matlab speech recognition code to get the desired recognition results. In this paper, the operating system for Linux systems in embedded hardware platform friendly arm mini2440 development board development board sound card resources and LED resources, and debugging the serial port. Prepared audio driver / dev / dsp and / dev / mixer, which correspond to the two parts of the audio driver, one is mainly used for recording broadcasts, and the other is used to control the size of the sound. Driver compiled successfully, and then loaded into the system kernel, and audio test program to test the two devices. In the development and application of test article is controlled through voice LED lamps, LED lights will be the type of voice signal in a different state, in order to achieve the purpose of the voice control embedded device. Finally, the application of voice-controlled airport ground services, and did some research of speech recognition technology to lay the foundation for further study.

Related Dissertations
More Dissertations