Research Of Embedded Speech Control Technology Based On μ'nSP^TM Processor

Development of embedded speech recognition controller has a significant for the man-machine communication and physical disabilities. At present, the hardware platform of speech recognition has not only computer, but also DSP chips. However, DSP chips have high price and difficulties in promotion and application of the shortcomings. So, this paper usesμ'nSP^TM family microcontrollers that have a low-cost functions like DSP for the hardware platform, and designs a small vocabulary and isolated word embedded speech recognition system.To realize the target, this article analyzes and summarizes the development of speech technology at home and abroad, and speech analysis processing algorithms and hardware platform in embedded field. The paper describes the speech signal processing technology, mainly studies cepstrum analysis and linear prediction analysis, which are related to the speech signal coding and identification, and the main program is given in the appendix. The paper discusses waveform coding, parametric coding and hybrid coding three voice compression technology. For the voice recognition technology, this paper mainly discusses the template matching and hidden Markov model.In terms of hardware design, the paper usesμ'nSP^TM MCU that has functions of DSP and cost-effective as core processor, and gives minimum system, power management unit, audio input and output unit, extern memory, communication and display, motor control, external clock, and so on for processor.Based on hardware platform, the paper develops the appropriate procedures for speech recognition controller. In the program development process, the paper takes voice signal acquisition, compression and recognition as key process, uses waveform coding and hybrid coding to complete voice coding, and uses the method of Hidden Markov Model (HMM) to complete speech signal recognition.At last, the experiment confirms the utility of this voice controller, this controller can better complete voice signal encoding and decoding, and recognition rate is up to 93% or more.