一种用于方言口音语音识别的字典自适应技术

Pronunciation Dictionary Adaptation Based Accent Modeling for Large Vocabulary Continuous Speech Recognition

基于标准普通话的语音识别系统在识别带有方言口音的普通话时,识别率会下降很多。针对这一问题,论文介绍了一种“字典自适应技术”。文中首先提出了一种自动标注算法,然后以此为基础,通过分析语音数据,统计出带有方言口音普通话的发音规律,然后把这个规律编码到标准普通话字典里,构造出体现这种方言发音特征的新字典,最后把新字典整合于搜索框架,用于识别带有该方言口音的普通话,使识别率得到显著提高。

It is well known that speaker variability caused by accent is an important factor in speech recognition，Aiming at this problem,a technique of modeling accent-specific pronunciation variations through pronunciation diction aryadaptation is presented.The paper firstly introduces a method of retranscribing at the phone level some accent specific data.The preferred transcription for each word is then compared to its dictionary entry and a list of phone replacement rules is generated.Using these rules to expand the canonical pronunciation dictionary,makes it be able to reflect the accent-specific pronunciation variations.At last,the new dictionary is integrated into the recognition framework to have its performance improved。