论文部分内容阅读
为了解决传统特征在重音检测中鲁棒性不高的问题,根据重音的定义,以单词为单位,考察词内各音素携带基音同步动态短时帧能量的差异,同时引入非线性加权因子,提出非线性加权能量特征。使用非线性加权能量特征以及与传统特征的特征组合对英语连续语音的实验结粜表明,非线性加权能量特征比传统特征鲁棒性更高,联合使用新特征与传统特征,可使系统误识率下降3.58%。
In order to solve the problem of low robustness of traditional features in accent detection, according to the definition of accent, the difference of the energies of the short-frame dynamic synchronization of the phonemes carried by each phoneme in each word is examined by word units. At the same time, the nonlinear weighting factors are introduced, Nonlinear weighted energy features. The experimental results of using nonlinear weighted energy features and the combination of features with traditional features on English continuous speech show that nonlinear weighted energy features are more robust than traditional features and the combination of new features and traditional features can make the system misunderstood The rate dropped 3.58%.