基于HTK的藏语拉萨话大词表连续语音识别的研究与实现

Recognition research and Realization of Large Vocabulary Continuous Speech in Tibetan-Lhasa Dialect Based on HTK

在分析藏语拉萨话特点的基础上,确定拉萨话音素集并制定了面向语音识别的拉丁转写方案.根据藏语字音转换规则,建立以单音节为单位的拉萨话发音字典,以音素为建模单元,训练得到三音子连续隐马尔可夫(HiddenMarkov Model,HMM)模型,探索藏语拉萨话语音识别的方法和总体框架,实现了基于HTK的藏语拉萨话特定人大词表连续语音识别.

Following the analyses of the characteristics of Tibetan, the phonemes set of Lahsa dialect and Latin phonetic transfer plan were determined according to Tibetan words-pronunciation conversion rules. A pronunciation dictionary was developed. The phoneme was took as a modeling unit, the contextual Hidden Markov Models was established. The framework of auto speech recognition of Lhasa dialect of Tibean was designed. Sperker - dependent Large - vocabulary continuous speech recognition of Lhasa Tibetan based on HTK was realized.