Lingyan QiaoQiaolan TangXinyu Sun
With the continuous popularization and development of the Internet, digital music resources continue to increase, how to effectively organize and manage these huge music resources, and explore related music resources effectively, in order to achieve the purpose of fast and accurate retrieval of these information, is an urgent problem in the field of music resource retrieval. The existing emotion recognition methods, no matter the single word or single sound emotion recognition methods, have their own shortcomings. In view of this, this topic proposes to effectively integrate the feature extraction capability of convolutional neural networks with the time series information of LSTM. On this basis, by studying the attention mechanism of CNN-LSTM, a single pattern recognition method based on CNN-LSTM is constructed for emotion recognition of sounds and music. Then, a composite neural network which can support the two types of feature information is constructed to improve the classification accuracy through the effective fusion of the two types of feature information. Compared with SVM, CNN, LSTM and other methods, this method has a great improvement, its accuracy reaches 68%, and the lyrics accuracy reaches 74%.
Nandkishor NarkhedeSumit MathurBhaskar AnandMukesh Kalla
Satish ChaurasiyaShubhangi Upadhyay
Caiyu SuJinri WeiMo YiWu Shanyun