摘要
语音分类是语音信号处理的重要组成部分。准确快速地对语音进行分类在语音编码、语音合成中有着重要的意义。针对当前一些常用分类特征和分类算法的不足,本文提出一种利用语音的Mel频率子带能量作为分类特征,建立高斯混合模型(GMM),运用最大后验概率准则对清音、浊辅音、元音分类的算法。仿真实验表明,在噪音环境下该算法仍可准确进行语音信号分类。
Speech classification is an important research topic in speech signal processing area. Rapid and precise speech classification is meaningful for speech coding and speech synthesis. Aiming at the deficiency of currently available classification features and classification algorithms, this paper proposes a novel algorithm through using the energy distribution within each frequency band in Mel-frequency scale as the classification feature and creating Gaussian mixture model and classifying the speech signal into voiced consonant, vowel and voiceless parts with the maximum a posterior probability. Simulation shows that the proposed algorithm is able to provide accu- rate classification result even in noisy environment.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2009年第9期1950-1955,共6页
Chinese Journal of Scientific Instrument
关键词
语音分类
能量分布
高斯混合模型
最大后验概率
speech classification
energy distribution
Gaussian mixture model
maximum a posterior probability
