基于藏语拉萨语语音识别中端点监测算法的研究

藏语语音识别技术研究已成为当今一大热点,而端点检测则是语音识别预处理阶段的关键环节之一。文章以标准拉萨藏语为对象,通过将加窗、短时能量和过零率相结合的方法对藏语语音识别技术中端点检测的算法进行了研究。实验表明,采用上述三种算法相结合的研究方法对浊音较重的藏语语音识别较为合适,可显著提高识别率。

Research on Endpoint Detection Algorithm in Speech Recognition based on the Lhasa Dialect of Tibetan Language

Research on Tibetan speech recognition technology has become a hot research topic of today. Theendpoint detection is one of the key parts in speech recognition preprocessing stage. In the present paper, theendpoint detection algorithm of Tibetan speech recognition was discussed based on Lhasa dialect of Tibetan lan-guage by windowing combining with short-time energy and zero crossing rate methods. The results of this studyshowed that the combination of the three algorithms is more appropriate method for recognizing the heavier so-nant Tibetan speech and obviously improving the recognition rate.