The Design And Implementation Of Old People Speech Emotion Recognition System

With the problem of aged tendency of population and empty-nest aggravating,the psychological problems of the elderly living alone have aroused wide public concern.Old people living alone do not usually live with children together,leading to the lack of emotional communication.What's more,there exist lots of problems for most of the intelligent pension system,such as low intelligence and poor emotional care.To deal with the above problems,based on the speech emotion recognition technology,this thesis designs and implements the old people speech emotion recognition system,which can intelligently sense the elderly's emotional information and automatically push emotional states,providing high-quality and personalized care service for the elderly.According to the system requirement,the thesis proposes an overall system scheme based on the C/S software development framework.Aiming at the key issues that need to be solved in the system,the relevant algorithms are studied and then applied in the system.Then function modules are designed in detail.The main work of this thesis is as follows:1.The thesis designs the overall structure into three layers: data acquisition layer,data processing layer and application layer.The data acquisition layer is responsible for collecting voice signal of the elderly.Then the voice signal is uploaded to the data processing layer and processed for speech emotion recognition.The application layer provides interactive platform for the elderly,children and backstage administrators.2.Since the elderly voice collected by the system is noisy,in order to extract pure speech signal and improve speech quality,the thesis presents an improved Wiener filtering algorithm based on prior signal-to-noise ratio.The improvement of the algorithm includes the estimation and correlation of noise power spectrum and the correlation of spectrum gain function.The simulation results reveal that the improved algorithm is superior to traditional algorithm in suppressing speech background noise.Then the CNNbased speech emotion recognition algorithm is improved of emotion recognition for pure voice.By using multi-scale convolution kernel,CNN can fully extract the feature information.The multi-resolution spectrogram and corpus fusion performance are compared and analyzed through experiments.The results indicate that our speech emotion recognition model gains better effect.3.The software system is designed and implemented in detail.The system consists of three parts: backend server,the elderly client and children client.The backend server is responsible for user management and speech emotion recognition which needs a large amount of calculation.The elderly client is mainly for old users,and responsible for voice acquisition and interaction with children.Children client is mainly for the elder's children,and responsible for interaction with parents and check the elderly's emotion status.Finally,in this thesis,the system is tested in detail.The test plan and content are designed,and the lab test platform is constructed.Based on the test platform,the system function is tested and analyzed,and the performance test is completed.The test result shows that the system can run stably and reliably,and can satisfy the desired functional requirements and non-functional requirements.