情感语音数据库优化及PAD情感模型量化标注

发布时间：2018-04-22 12:12

本文选题：情感语音数据库 + 维度情感描述　；参考：《太原理工大学学报》2017年03期

【摘要】：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。
[Abstract]:Emotional speech database is the basis of emotional speech recognition. It is of great significance to establish a dimension emotional speech database including cognitive psychological factors to improve the recognition rate and the human-computer interaction ability. In this paper, the author first selects and optimizes the TYUT2.0 database based on the recognition rate threshold. The emotional speech database consists of four kinds of emotional sentences. Of them, 62 are sad, 58 are angry, 57 are happy, 60 are surprise. Then the 3D emotion model of PAD is used to label the voice of the database, and the dimension emotional voice database is obtained. Each sentence in the database has a corresponding audiological identification rate and PAD value. The PAD value of each sentence is statistically analyzed, which proves the validity of the emotional speech database of this dimension, and lays a foundation for the study of dimension emotion recognition in the future.
【作者单位】：太原理工大学信息工程学院;
【基金】：国家自然科学基金资助项目(61376693)
【分类号】：TN912.3;TP311.13

【相似文献】