四川口音普通话的声学建模

Acoustic Modeling for Sichuan Accent

考虑到对因受地方口音导致的相近、相似、模糊发音的语音的识别困难,本文以易受四川口音(含有特定的声母zh-z/ch-c/sh-s)孤立词语音为识别语音,对其进行声学建模。通过对初始的49维特征参数分三个步骤进行二次特征提取,得到对特定四川口音敏感的特征参数。最后采用矢量量化技术对初始特征参数和通过二次特征提取得到的特征参数进行对比实验。实验表明,经过二次特征提取得到的特征参数有效地改进了易受四川口音(含有特定的声母zh-z/ch-c/sh-s)影响的语音的识别效果,提高了识别率。

In consideration of the difficulties of recognizing the inaccurate,likeness,misty of pronounce being subjected to Sichuan-accent influence,the paper choose those speech being subjected to Sichuan-accent influence whose initial consonants are zh,z,ch,c,sh and s as experimental object for carrying acoustic modeling.The paper extraction the sensitive characteristic parameters for Sichuan-accent through extracting secondary feature parameters for original characteristic parameters.Finally,the paper will make out comparison for original characteristic parameters and secondary feature parameters adopting Vector Quantization and the experiment expresses that secondary feature parameters improve recognition effect of the speech being subjected to Sichuan-accent influence and increase recognition rate.