| In the recent twenty years,speech recognition technology has made great progress and is gradually from the laboratory into the market. Experts predict that speech recognition technology will enter the fileds of industry, home appliances, communications,automotive electronics,medical,consumer electronics etc in the next 10 years.Although the current speech recognition technology has made great progress, but there is great research value in the ways of larger words,high real-time,the system of low resource consumption and high recognition rate.In daily life,speech technology has great application and development in the field of service robot control, speech control of service robot motion has value and wide application prospect.This paper introduces the voice control system that uses speech recognition technology to control the motion of service robots.First of all,this paper introduces the research content of historical background and research status at home and abroad, including the analysis of the research status of service robot control, speech recognition,speech endpoint detection and speech feature dimension reduction.Secondly,this paper introduces the method for pretreatment of voice data,Including speech signal preemphasis framing,windowed and endpoint detection.This chapter puts forward the endpoint detection method based on logarithmic energy spectral entropy,because of its simple calculation,it can be applied to the low-end embedded platform and laid the foundation. for the control program that would been ported to the Android mobile phone.Then, this chapter introduces the method of extracting Mel Frequency Cepstral Coefficients of speech(MFCC),MFCC provides the voice signal feature data forspeech recognition.Third, this chapter introduces the three algorithms of hidden Markov model(HMM) and Compares the advantages and disadvantages between Dynamic Time Warping (DTW) algorithm and hidden Markov model(HMM). Because the speech control system of service robot is the object of speech recognition for the actual situation of non specific people,this paper selected the hidden Markoff model(HMM) as the speech recognition algorithm of speech control system of service robot.Fourth,because the speech feature parameters have the problems of redundant data and high dimensionality, resulting in the training of hidden Markov model for long time, this paper decide to reduce the dimension of the speech feature parameter(MFCC) by using the method of principal component analysis(Principal Component,Analysis,PCA). because the system calculation is too large and too much memory, leading to too long identify time in the process of large vocabulary speech recognition,this paperpresents the K mean clustering algorithm based on speech feature parameters, the lower dimensional(MFCC)K mean clustering in order to get the characteristic parameters of stable block.Fifth,the service robot voice control system of design is introduced in this chapter, the development of this system is ton the eclipse platform by use of JAVA language programming,the first part is mainly to develop the control interface of system,he second part is develop the program of voice data processing and speech recognition algorithm.Sixth,this chapte introduce the simulation and data analysis of algorithms and manual debugging service robot voice control system.The experimental results show that the endpoint detection method based on logarithmic energy spectral entropy Improves the accuracy of endpoint detection.In order to ensure the accurate rate basically unchanged,by dimensionality reduction of speech feature parameters by PCA and speech template group,the system greatly reduces the speech training time, improves the real-time speech recognition,and achieved good recognition effect.It studys the speech endpoint detection method based on logarithmic energy spectral entropy nd reducing the dimension of the speech feature parameters by using the Principal component,it also studys grouping method for low dimensional data using Kmeans clustering. |
PDF Full Text Request