基于声门特征参数的语音情感识别算法研究

摘要

In order to implement an efficient automatic speech emotion classification system,a speech emotion detection algorithm is proposed using glottal extraction features and Gaussian mixture models.The proposed algorithm is based on human speech production model.The glottal signals are calculated using inverse filter and linear prediction analysis.The glottal time domain features are then extracted to distinguish various emotion types.The publicly available BES (berlin emotion speech database) is used to classify seven different emotions:angry,bored,disgust,fear,happy,neutral and sad.The experimental results show the proposed algorithm achieve better performance on the emotions detection than traditional fundamental frequency and formants parameters,and it is close to human decision.%为实现更为有效的自动语音情感识别系统,提出了一种基于声门信号特征参数及高斯混合模型的情感识别算法.该算法基于人类发音机理,通过逆滤波器及线性预测方法,实现声门信号的估计,提取声门信号时域特征参数表征不同情感类别.实验采用公开的BES (berlin emotion speech database)情感语料库,对愤怒、无聊、厌恶、害怕、高兴、平静、悲伤这7种情感进行自动识别.实验结果表明,提出的语音情感识别系统能有效的识别各类情感状态,其情感判别正确率接近于人类识别正确率,且优于传统的基音频率及共振峰参数.