论文部分内容阅读
在语音识别中，为了得到分布共享的异音模型，先要知道与发音语境无关的音素模型．在本文中，给出一种用于训练与发音语境无关音素模型的方法，然后利用这种音素模型完成对异音模型的训练、以及对异音模型的输出分布的二值决策树聚类．实验结果表明，使用给出的方法，可以实现对与发音语境的无关音素模型，以及异音模型的可靠训练．
In speech recognition, in order to obtain heteronuclear model of distribution and sharing, it is necessary to know the phoneme model irrelevant to pronunciation context. In this paper, we give a method to train the irrelevant phoneme model related to the context of pronunciation, and then use the phoneme model to train the alphanumeric model and the binary decision tree clustering of the output distribution of the alphanumeric model . The experimental results show that the proposed method can be used to reliably train irrelevant phonemes related to pronunciation contexts and abnormal noise models.