离散标签与维度空间结合的语音数据库设计

Design of discrete tags and dimensional space combined emotional speech database

建立了一个将离散情感标签与维度情感空间结合起来的汉语情感语音数据库。该数据库由16名母语为汉语的说话人对情感语音进行表演型录制。语音样本是根据中性、愉悦、高兴、沮丧、愤怒、哀伤,以及悲伤等七种离散的情感标签采集而得,每名说话人有336条语音样本。随后由三名标注人在维度空间上对每条语音样本进行标注。最后,根据标注所得的数据来研究这七种情感在维度空间的分布情况,并分析了情感在一致性、集中性和差异性方面的性能。除此以外,还计算了这七种情感的情感识别率。结果显示,三名标注人对该数据库标注的一致性都达到了80%以上,情感之间的可区分度较高,并且七种情感的情感识别率均高于基线水平。因此,该数据库具有较好的情感质量,能够为离散情感标签到维度情感空间的转化提供重要的研究依据。

This paper establishes a Mandarin emotional speech database that combines discrete emotion tags with dimensional emotion space. The database is recorded for 16 Chinese native speakers in performing Chinese emotional speech. The speech samples are acquired from seven discrete emotion tags, such as neutrality, pleasure, happyness, frustration, anger, sorrow, and sadness. Each speaker receives 336 utterances. Then, each of the speech samples is annotated by three annotators in dimensional space. Finally, according to the obtained data, the distributions of these seven emotions in the emotion space are studied, and the performances in consistency, concentration and difference of these emotions are analyzed. Besides, we calculate the emotion recognition rates of these seven emotional speech. The analyses show that the consistencies of the three annotators for the database are more than 80%, and these emotions can be distinguished, in addition, the recognition rates of these seven emotions are all higher than baseline level. Therefore, the database has a good emotional quality, and can provide important research basis for the transformation of discrete emotion tags to dimensional emotion space.