摘要
提出一种基于改进GMM模型的耳语情感语音识别方法。该方法在GMM的每个成员通过用矢量量化误差值取代传统GMM的输出概率值来计算模型的得分,使得建模时所需训练数据量减少,并且识别速度有所提高。实验结果表明当训练数据较少时,提出的新的识别方法的实验结果明显好于传统的GMM方法,证明了该方法的有效性。
A new whisper emotional speech recognition algorithm based on improved GMM is proposed in this paper.In this algorithm,every member in GMM utilises the vector quantisation error scale instead of the probability value of the output used in tradition GMM to calculate the score of the model,thus reduces the training data amount when modelling and improves the recognition speed.Experimental results show that when the training data is lesser,the new recognition algorithm proposed has an obviously better experimental result than the tradition GMM,which verifies the effectiveness of the algorithm.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第11期73-74,136,共3页
Computer Applications and Software
基金
国家自然科学基金项目(60872073
60975017
51075068)
教育部博士点专项基金(20110092130004)
江苏省高校自然科学基金项目(10KJB510005)
关键词
耳语语音
高斯混合模型
情感识别
Whispered emotion speech Gaussian mixture model（GMM） Recognition
