摘要：The proposed work combines the evidence from mel frequency cepstral coefficients (MFCC) and residual phase (RP) features for emotion recognition in music. Emotion recognition in music considers the emotions namely anger, fear, happy, neutral and sad. Residual phase feature is an excitation source feature and it is used to exploit emotion specific information present in music signal. The residual phase is defined as the cosine of the phase function of the analytic signal derived from the linear prediction (LP) residual and also it is demonstrated that the residual phase signal contains emotion specific information that is complementary to the MFCC features. MFCC and residual phase features are used to create separate models for each emotion. The evidence from the models is combined at the score level for each emotion and it is used to recognize the emotion. The proposed method is evaluated using music files recorded from various websites and the method achieves a performance of 96.0%, 99.0%, 95.0% using AANN, SVM, RBFNN, respectively.
关键词：Music emotion recognition
; Mel frequency cepstral coefficient
; Residual phase
; Autoassociative neural network
; Support vector machine
Loading...
