情感语音数据库优化及PAD情感模型量化标注

Emotional Speech Database Optimization and Quantitative Annotation Based on PAD Emotion Model

情感语音数据库是情感语音识别研究的基础,建立包含认知心理因素在内的维度情感语音数据库对提高识别率、改善人机交互能力具有重要意义。笔者首先对前期建立的摘引型TYUT2.0数据库进行语音听辨筛选,根据认同率阈值进行数据库优化,得到的情感语音数据库包含四种情感的语句237句,其中"悲伤"62句,"愤怒"58句,"高兴"57句,"惊奇"60句。然后利用PAD三维情感模型对该数据库语音进行标注,得到维度情感语音数据库。该数据库中的每句语音都有对应的听辨认同率以及PAD值。对每句语音的PAD值进行统计分析,证明了该维度情感语音数据库的有效性,为今后研究维度情感识别奠定了基础。

Emotional speech database is the foundation of emotional speech recognition research,it has great significance to establish a continuous dimension emotional speech database including cognitive psychological factors for improving the performance of the speech emotion recognition and human-computer interaction.In this paper,first,hearing screening was conducted on previously established TYUT2.0database,then the database was optimized according to recognition rate threshold.The resultant emotional speech database with 237 speeches has four types of emotion including 62,58,57,and 60 speeches representing respectively sadness,anger,happiness and surprise.The speech of this database was marked by using PAD emotion model,giving a dimensional emotion database.Each speech has its identification rate and PAD value.Statistical results of PAD value prove the validity of this dimensional emotional speech database,which lays the foundation for studying emotional speech recognition in continuous dimension in the future.