Deep Emotion Recogntion Of The Music By Mixd Voice And Lyric

With rapid development of computer science and technology for the past 20 years,the artificial intelligence has become the hottest field in computer science and technology.The affective computing is an important constituent part of artificial intelligent science.Through affective computing,it enables machine to study and simulate human's emotions,which has become an important objective for the development of artificial intelligence.The voice emotion is an important research direction of affective computing.However,music is an important method for people to express emotions in life,which is the phonological representation form that could mostly affect emotional influences as music contains plenty of emotive information.After research,it finds that song's melody not only includes music's emotive information,it also includes emotive information in song's lyrics.The combination between song's melody and lyrics can reflect a complete emotional characters.Regarding the classification of song's emotion,this article points out an in-depth emotion analytical method based on multi-moding two-times integration.This article conducts analysis of music in the aspects of voice frequency and lyrics.It respectively extracts two different audio features MFCC and MIDI as well as the textual features of lyrics.After integration of two different features of audio's MFCC and MIDI,it conducts the second integration between audio's integration feature and textual features of lyrics.With this method,the features of emotional integration for songs are generated after integration for two times.After acquiring song's features of emotional integration using two-times integration,it optimizes and classifies song's features of emotional integration using Restricted Boltzman machine and Convolutional Neural Networks.In addition,the training and testing are conducted on corpus of song's emotions.The multi-group experiments indicate that the musical emotion can be acquired based on the multiplex-mode in-depth learning model,thus to classify the emotions.A more accurate musical emotion can be judged,which can deepen the reflection of emotive information and tendency contained in music.