Research On Infant Emotional Information Recognition Method Based On Infant Cry Detection

In recent years,as the ageing of the population has become increasingly serious and employment pressures have gradually increased,more and more female compatriots have moved to work positions and are unable to effectively accompany infants.However,the sense of security gained in the first few months after the baby's birth will be accompanied by life in the future,so meeting the baby's needs in time will al ow the child to grow healthily.At present,the way in which the elderly and the nanny are used instead of parents is obviously insuffic ie nt in understanding the needs of the baby,especially when the baby cries.In the field of infant crying emotional needs information,the vast majority of research work has focused on finding effective features for infants in different emotional needs and identifying different emotional needs in infant cries.However,since the environment in which the baby is located is not an absolutely quiet environment,the environmental noise will greatly reduce the recognition rate of infant crying emotional need information.Therefore,how to improve the robustness of information recognition models for different emotional needs in infant crying is a problem that needs to be solved urgently.Based on the existing research,this project plans to carry out research work in the following areas:(1)Established a crying voice corpus of infants with different emotional needs informat io n.The corpus is the basis for the study of speech emotion recognition.At present,there is no unified corpora at home and abroad.In view of this situation,this paper constructs a corpus of infant crying emotional needs information.(2)The anti-noise robustness of different emotional needs information recognition models in infant crying is studied.In this paper,the combination of prosodic features and related spectral features is used as the original feature.The anti-noise robustness of Softmax regression algorithm,artificial neural network and support vector machine are studied.Experime nt a l results show that although all three recognition models have certain robustness,the effect is not ideal.(3)An improved convolutional neural network is proposed to enhance the robustness of different emotional needs information recognition in infant crying.In this paper,on the basis of traditional convolutional neural networks,a convolutional neural network based on multi-sca le convolution kernels and multi-pooling method is proposed to enhance the robustness of different emotional needs information recognition in infant cries.Experiments show that the improved convolutional neural network has achieved good results in enhancing the robustness of the identification of different emotional needs in infant crying.(4)The robustness of different emotional needs information in infant crying is proposed based on the multi-scale local binary pattern feature of Gabor gray map.On the basis of the traditional local binary pattern of grayscale image spectrum,first generate the spectral image grayscale image;secondly,use the Gabor wavelet filter to filt;finally use the multi-scale block local binary pattern to extract features of the image.Experiments show that the feature of mult iscale block local binary pattern based on Gabor gray-scale maps is good for enhancing the robustness of information recognition for different emotional needs in infant crying.Compared with traditional features,its anti-noise performance is better.