【作者】 吴伟民； 林道发； 杨家沅；

【Author】 Wu Weimin Lin Daofa and Yang Jiayuan (Sichuan University)

【机构】 四川大学；

【摘要】 我们用LPC倒谱,△倒谱,归一化能量作特征,以汉语无调音节为识别基元,采用连续密度HMM,引入音长概率,对多人的汉语连续语句进行识别,获得初步结果。训练与识别算法分别为分段K-means算法和帧同步网络搜索算法。在这时我们提出了改进识别性能的若干设想。更多还原

【Abstract】 In this paper a speaker - independent continuous Chinese speech recogntion system is discussed. In the systum, syllables disregarding tones are the recognition units, Continuous Densities Hidden Markov Models (CDHMM) and syllable duration probability are incorporated in. Segment K-means algorithm and frame - synchronous network search algorithm are used to train and recognize respectively. Several methods are presented to improve the performance of the recognition system.更多还原