A speech emotion recognition method in cross-languages corpus based on feature adaptation

For speech emotion recognition on cross-corpus, we study the problem of speaker feature adaptation. First, we discuss the existing approaches in adaptive emotional classification from speech signals. Second, the speaker feature adaptive approach is further studied in view of additive emotion feature distortion. Finally we verified our approaches using different cross-languages corpus, including German, English, Chinese and Vietnamese. The experimental results demonstrate an improved speaker feature adaptive ability especially when a large number of unknown speakers are presented. Therefore, in this speaker independent scheme we can demonstrate the ability of incremental learning character of the proposed method.