摘要： 为获得更丰富的情感信息、有效识别长语音的情感状态，提出基于D-S证据理论的多粒度语段融合语音情感识别方法。采用2种分段方法对语音样本分段，用SVM对语段进行识别，再利用D-S证据理论对各语音段识别结果进行决策融合，得到2种分段方法下语音的情感识别结果，将这2个识别结果进一步融合得到最终结果。实验结果表明，该方法具有较好的整体识别性能，能有效提高语音情感的识别率。

关键词: 语音情感识别, 支持向量机, D-S证据理论, 语句分段, 决策融合

Abstract: To obtain more abundant emotional information and recognize emotion of long speech effectively, a new method for speech emotion recognition of multi-granularity segment fusion using D-S evidence theory in decision fusion is proposed. Speech sample is segmented by two methods. Each segment has a result after being recognized by SVM. Results of segments carry through fusion by D-S evidence theory to get those of speech samples. Results of two segment methods fuse again. Experimental results show that, the recognition performance of this method is better, and the speech emotion recognition accuracy rate is improved effectively.

Key words: speech emotion recognition, SVM, D-S evidence theory, utterance segment, decision fusion

中图分类号: 

• TP391.42