Abstract:

To improve efficient utilization of local features in sentiment words of speech, we propose a new speech emotion recognition method by fusing local features of sentiment words and global features. Based on the acoustic features of the emotional dictionary, we extract local features, such as sentiment word existence in the statement and its density. Then by fusing the global features of speech, we can recognize the speech emotion through modeling with machine learning algorithms. The comparative experiment results show that the proposed method can gain very good performance, and the introduction of local features can improve the precision of speech emotion recognition.

Key words: speech, emotion recognition, speech emotion dictionary, local feature, global feature