Research On Human Voice Color Classification Based On Deep Neural Network

With the popularity of application services such as Internet voice social networking and voice assistants,people usually have their own unique preferences for the timbre of voice content.Specifically,the application service provider can classify the voice data of different users by timbre,and recommend content of a specific timbre to different users according to the user's preference.Using human voice classification technology can significantly improve the user experience,increase the number of users and improve the user retention rate.Therefore,the study of human voice color classification has important practical value,and there are few related studies about it.This paper mainly studies the timbre classification of human voice,that is,the classification and recognition of human voice timbre.This is the first work for human voice timbre classification.This paper mainly uses the deep neural network to carry out the technical research and system realization of human voice color classification.First of all,based on the original voice data provided by the project partner company,this paper provides a large-scale real human voice and voice data set through data preprocessing and other methods,which solves the research status of the lack of available data sets for human voice and voice classification.Based on this data set,this paper implements a human voice color classification model based on convolutional neural network,and selects the best feature combination as the input feature of the model through experiments.The experimental results show that the model has good vocal color classification ability.Secondly,this paper implements a human voice color classification model that integrates the least mean square noise reduction algorithm and convolutional recurrent neural network,pre-processes the input audio through different noise reduction algorithms and compares the performance of the model.The experimental results show that after the noise reduction process,the proposed human voice color classification model has achieved a good effect of human voice color classification,which can effectively improve the classification accuracy of the model.It has good generalization on the dataset.Finally,based on the above research on human voice and tone classification technology and model,this paper designs and implements a prototype system for human voice and tone classification,which integrates the method studied in this paper.Users can upload their own recordings in the system or directly in the browser.Record and identify the corresponding tone.