藏语孤立词语音识别系统研究

藏语语音研究相当滞后,文章结合语音识别知识和藏语特点,尝试性地研究了藏语孤立词语音识别研究.首先提取MFCC参数作为语音特征参数,形成语音模板库,采用DTW模型实现了语音识别系统,并且针对藏语孤立词多音节的特点,改进了传统的基于短时能量和短时过零率双门限检测法,即加入了音节间静音段时长门限,提高了孤立词语音信号检测的准确性和识别率.

Research on Tibetan Isolated-word Speech Recognition System

So far, the study of Tibetan speech has not reached higher level. By combining the knowledge of speech recognition and feature of Tibetan, Tibetan speech recognition were explored and studied through regarded Tibetan as isolated words in the present thesis. The paper considered Tibetan as isolated word, which with polysyllable feature, improved old detective method, double - threshold detection method that is based on short - time energy and zero - crossing rate, the new detection method has been introduced length of mute stage among syllables, the method can improve speech signal detect of isolated words in both sides, namely accuracy and recognition rate.