Research On A Speech Recognition System For NAO Robots

Researching and realizing the intelligentization of robots and using more intelligent service methods to meet the needs of social development are of great significance to improving the quality of human life.Researching the robot speech recognition system to achieve human-computer interaction is the precondition for intelligent robots to provide intelligent services.As one of the most widely used robots for scientific research in the world,NAO robot has important significance and value for the research of NAO robot related speech recognition system.In order to realize the intelligent interaction between human and NAO robots,a co-simulation platform for NAO robot-based speech recognition system was designed.Firstly,the development and research status of speech recognition are analyzed.The basic principle and signal processing technology of speech recognition technology are studied in depth.Three endpoint detection methods are studied: short-term energy,short-term zero-crossing rate,and double threshold method.The principle of the double-threshold method is analyzed in depth;the three feature extraction techniques of speech signal: linear prediction coefficient(LPC),linear prediction cepstrum coefficient(LPCC)and Mel spectrum cepstrum coefficient(MFCC)are analyzed in depth.In addition,the feature parameter comparison experiment was carried out for the two feature extraction methods of the linear prediction cepstrum coefficient and the Mel spectrum cepstrum coefficient.By comparison,the feature extraction method based on the Mel spectrum cepstrum coefficient has higher matching degree.Secondly,the speech recognition models and algorithms are summarized and summed up.The DTW algorithm and the Hidden Markov Model(HMM)model are deeply studied.The basic principles of the DTW algorithm are analyzed,and the research is based on the whole.The DTW algorithm experiment of path constraint and search width limitation is improved.The effectiveness of the two improved algorithms is verified by comparing the improvement rate of the algorithm before and after the recognition.At the same time,the contrast experiments of the DTW algorithm based on the two feature extraction methods of LPCC and MFCC are used respectively.It is concluded that the DTW algorithm based on Mel spectrum cepstrum coefficients has a high recognition rate,which shows that Mel spectrum cepstrum coefficients can reflect the characteristics of speech signals to a certain extent.Three problems and their solutions to the HMM model are studied in depth.The common problems in the solution of the HMM model such as selection of initial values,insufficient training data overflow problems,and training of multiple observation vector sequences are analyzed.The HMM model is established.The speech recognition system has determined the best state value of the HMM model through experiments,the higher the signal-to-noise ratio is,the better the system recognition effect is,and the better conclusion of the HMM model in identifying continuous words than the DTW algorithm.Finally,a co-simulation platform of NAO robot speech recognition system based on HMM model was built.Summarizes the process of acquisition,processing,transfer,machine training,and recognition of audio by the co-simulation platform;through the establishment of the transmission control protocol(TCP/IP protocol),it is completed between the NAO robot as a server and the client's MATLAB Communication.It realizes the capture of audio by the server,the processing and recognition of the voice information by the client,and transmits the recognition result through the protocol to Choregraphe the control software of the server.The control software triggers the process of the specific behavior instruction according to the recognition result.The co-simulation platform is simple in structure,easy to expand and transplant,and has better recognition performance.