摘要
高斯混合模型在语音转换中得到了广泛应用,但其随着模型阶数和特征维数的提高,估计参数的数目会急剧增加,使参数估计的准确性和稳定性大为降低。本文提出将特征向量去相关之后,将向量之间的转换转化为标量之间的转换,以此来减少估计参数个数。实验表明,采用该方法的语音转换算法能有效改善转换语音的性能。
GMM is one of the most useful algorithms in voice conversion.However with the increase of model ranks and dimensions of characteristic vector,the number of parameters to be estimated increases rapidly,which affects the precision and stability of the parameter estimation.This paper proposes a new method which translates vector transformation into scalar transformation.Experiment results show that this method can improve the converted voice＇s quality.
出处
《计算机与现代化》
2010年第9期82-84,共3页
Computer and Modernization
