The Research And Implementation Of Speech Recognition Technology For Smart Home

Smart Home is a typical set of computer, communications and consumer asone of the3C system，all of smart devices will be connected into a whole sensortechnology, network transmission technology, information processing technology,audio and video technology, using an efficient management system controls alldevices. Language communication is the most direct way information exchange.During the use of Smart Home system, combined the speech recognitiontechnology and control technology is becoming the focus of current research. Intraditional smart home system, people use a speech recognition chip to controlsmart devices, and that the existing speech recognition products depend on theInternet platform. In order to save hardware resources, to change dependingInternet, we propose embedded speech recognition software modules in the smarthome management system, intelligent speech control equipment.This thesis is based on speech recognition technology, a detailed analysis ofspeech recognition technology, the basic principles and implementation processes.From speech input to recognition result output, needs preprocessing, featureextraction, pattern matching and modeling four steps. In the preprocessing stage,improve the traditional dual-threshold endpoint detection in a multi-segmentextraction shortcomings; we propose a judgment based on double thresholdenergy endpoint detection algorithm is used to weaken environmental noise audiosegments. This algorithm used to extract different situations in the entiremulti-tone audio signal segment. In the speech signal feature extraction process,use the auto-correlation function for pitch detecting. Because of speechinformation concentrated in the low-frequency part and Mel-Frequency CepstrumCoefficient converted to the linear frequency Mel-frequency, highlighting the lowfrequency information of the speech signal, thus contributing to speechrecognition. In the pattern matching process, using dynamic time warpingrecognition algorithm to match the model library of voice messages, enablesvoice control smart home.This thesis has designed two speech recognition operation softwares: basedon the Android platform and Matlab development environment, comparing theadvantages and disadvantages of the two types of software. For Matlabdevelopment software and this system has been the recognition rate, real-timetesting. The results showed that the recognition rate of mono segment achievedgood recognition results prove that this system identification algorithm has good practicability. The inadequacies of the system lies in the multi-tone recognitionprocess segment recognition rate is not high, real bad, this article analyzes thespecific reasons, and proposed to modify the direction.