Research On Speech Emotion Recognition And Its Application In The Service Robot

Emotion plays an important role in Human-Machine interaction.Speech is one of the most widely used communication modes in people's daily life and an important tool expressing emotions.The aggravation of global aging,less children and "empty nesters" has provided a broad market prospect for the development of the service robot.With gradually profound new generation of Human-Machine interaction techniques,the application of speech emotion recognition in service robot has become a research hotspot.In the paper,three aspects of speech emotion recognition applications in service robot are carried out as follows:(1)In the field of machine learning,Zernike moment-based speech emotional feature extraction algorithm(ZMFCC)was proposed.Reliable recognition of emotion heavily depend on the extracted features because they represent the acoustic contents of speech.As one of spectrum features of speech signals,Mel-Frequency Cepstral Coefficients(MFCC)gives full consideration of auditory perception characteristics of human ears,so it has been extensively applied in related fields of speech as the research object.MFCC feature extraction algorithm was improved in this paper,MFCC speech emotional feature extraction algorithm based on Zernike moment was proposed,and LIBSVM classifier was combined to implement classification and recognition of six emotions in Chinese emotional corpus CASIA.The experiment shows that the proposed algorithm in this paper is superior to spectrum feature-based speech emotional feature extraction algorithms like MFCC and HuWSF.(2)In the field of deep learning,CNN-RF model based on combination of convolution neural network and random forest was proposed for speech emotion recognition.Deep learning can make automatic learning in the original data and extract deep features.In this study,CNN was used as the feature extractor and RF was used as the classifier,a network model based on combination of convolution neural network and random forest(CNN-RF model)was proposed for Chinese speech emotion recognition.Firstly,speech signals were transformed into spectrogram for normalization and input into the CNN to extract speech emotion features.Extracted speech emotion features were classified by the RF algorithm.The CNN-RF model was trained and tested on speech emotion database CASIA of Chinese Academy of Sciences,which was proved superior to traditional CNN model.(3)The NAO robot instruction box was improved,and the speech emotion recognition model was applied in the NAO service robot platform successfully.To keep format consistence between voice recorded by NAO robot and voice in CASIA,the Record Sound box carried in NAO robot was improved and the robot recognized people's emotions(happy,angry,sad and joy)from speeches and realized more intelligent Human-Machine interactions.