【作者】 秦勇； 莫福源； 李昌立； 关定华；

【Author】 Yong Qing Fuyuan Mo Changli Li Dinghua Guan (Institude of Acoustic,Academia Sinica)

【机构】 中国科学院声学研究所；

【摘要】 本文为了解决汉语大词汇语音识别中词汇的音调识别问题,提出了一种基于半连续隐马尔可夫模型的二字词音调模式的自动识别方法,通过对大量语音数据的观察分析,就特征矢量的提取、矢量量化技术中初始码书的获取和HMM的输出概率密度函数的定义等方面进行了深入研究,实验证明,该方法识别率为95％以上,顽健性强,是大词汇语音识别的一个重要方面。更多还原

【Abstract】 In this paper some initial efforts based on SCHMM in the recognition of the tone-pattern of Mandarin double-syllablic word have been made in order to get the goal of large vocabullary recognition, based on the anylysis of a great amount of speech datas which focused much emphasis on extracting the feature vectors,getting the initial VQ code book and defining our SCHMM ’s output probabilistic density funtion and so on. The recognition accuracy was found to be about 95% or higher,we believe that our efforts will be an important aspect of our large vocabullary speech recognition system.更多还原