越南语语音情感特征分析与识别

Emotional Feature Analysis and Recognition from Vietnamese Speech

本文研究了越南语语音情感识别问题,采集并建立了越南语情感数据库。母语为越南语的两名男性和两名女性说话人进行了越南语情感语音的表演和录制,通过多人听辨实验进行了数据的验证和评选,建立了一个初步的越南语语音情感数据库,为今后的跨语言研究提供数据基础。在所获得的情感数据上,提取了基本的语音特征,并构造了可用于情感识别的静态统计特征。采用高斯混合模型进行了建模与识别,识别结果显示,本文中的系统能够较好的识别越南语语音情感。在今后的工作中,可以进一步研究跨语言情感特征的分析与识别。

In this paper we studied the emotion recognition from Vietnamese speech signal,and established a Vietnamese emotional speech database.Two male subjects and two female subjects whose native language is Vietnamese participated in the acting of emotional speech.Through a listening test by multiple listeners the emotional data is selected and a basic Vietnamese speech emotion database is achieved,which may serve as a data foundation for future cross-language study.Based on the collected emotion data we extract the basic acoustic features and construct the static emotional features which are used for modeling and recognition.Emotion model is built and tested using Gaussian mixture models,and experimental results show that the emotion recognition system proposed in this paper successfully detects several basic emotions from Vietnamese speech.In the future work a further study on the cross-language emotional feature analysis and recognition is still needed.