Research On The Speech Interaction Technology Of The Elderly Escort Robot

As an indispensable member of the robot family,accompanied by a higher degree of intelligence,the escort robot can be widely used in the family,hospital,hotel,nursing home and other scenes.In recent years,it has developed rapidly at home and abroad.The project that was put forward by the school and the enterprise includes a series of robots,which are related about company and nursing.And modular design is used to meet different market needs.The escort robot for the elderly mainly aims to reduce the loneliness of the elderly and reduce the social pressure.Therefore,the design of an escort robot prototype system based on intelligent speech interaction needs to ensure the fun,safety and reliability for the elderly.Based on the actual needs of the enterprise,the paper designs and integrates the intelligent aged escort robot system based on speech recognition.In order to improve the purity of speech and the recognition performance of the system,such key technologies as signal enhancement,feature extraction and acoustic model construction in speech recognition are studied in detail.The details include the following aspects:(1)According to the actual project requirements and the background of the project,the overall system scheme of the robot is designed and the system structure of the escort robot based on speech recognition is analyzed.The detailed design and structure analysis of the whole system are made from such aspects as hardware system design,chassis design,shape design of the prototype and speech interaction module etc.And an intelligent escort robot frame based on speech recognition is constructed.(2)To solve the problem of low contrast between the energy of the words of the elderly and the energy of the environmental noise,the problem of speech enhancement is discussed,and the speech enhancement based on the joint maximum posterior probability is proposed.Aiming at different additive noise,the paper studies from the traditional spectral subtraction enhancement algorithm and improves the inherent defect of the algorithm.And a joint phase estimator based on the iteration of alternating phase and spectrum is proposed.Besides,the multiband spectral subtraction is introduced to eliminate the influence of "music noise".Finally,a good noise reduction effect is achieved.(3)The paper studied the feature extraction technology of speech recognition.In order to obtain the speech feature parameters with good distinction and robustness,the traditional MFCC feature extraction algorithm is taken as the research subject.And the weighted optimization based on the F ratio is carried out.Meanwhile,the feature parameters of different time domain and frequency domain are added to optimize the combination contrast experiment.Finally,the dimensionality of the optimal combination is reduced to improve the computational efficiency.It provides a preliminary research foundation for pattern recognition.(4)On the basis of feature extraction,in order to improve the performance of system recognition,the system performance comparison experiments of two classical models are completed,which is based on the Kaldi open source platform.And the recognition performance of the two systems based on the GMM-HMM model and the DNN-HMM model are studied respectively.Among them,the DNN-HMM model uses neural network instead of Gauss model to calculate state output probability in HMM model.(5)The paper designs the human-machine interface,integrates the robot intelligent interaction system and motion control system,which is based on integrated development environment on the host computer.The functions of the communication between modules,robot movement and obstacle avoidance debugging,the data analysis and processing of speech recognition and expression switching are tested.Finally,the desired design requirements are achieved after a lot of debugging.