首页>
外国专利>
VOICE RECOGNITION METHOD USING HIDDEN MARKOV MODEL HAVING DISTORTION DENSITY OF OBSERVATION VECTOR
VOICE RECOGNITION METHOD USING HIDDEN MARKOV MODEL HAVING DISTORTION DENSITY OF OBSERVATION VECTOR
展开▼
机译:使用具有观测矢量失真密度的隐马尔可夫模型的语音识别方法
展开▼
页面导航
摘要
著录项
相似文献
摘要
The present invention relates to speech recognition using distortion of an observation vector, and the distortion probability density during the learning process of a hidden Markov model (HMM) having a distortion probability density of the observation vector. (B) obtaining a state segmentation sequence; calculating an average length mi of each state i with reference to the state segmented learning observation sequence; and a mean vector υij for each normalized time interval j in the state i. Obtaining a distortion eijk for the average vector for each normalization time interval j in each state i for each learning observation sequence k; And obtaining a minimum value ( nij ) and a maximum value ( xij ) of distortions of the total k learning observation vectors for each normalized time interval j in each state i.;As described above, the present invention provides a better speech recognition rate than the general HMM.
展开▼
机译:本发明涉及使用观察矢量的失真的语音识别,以及在具有观察矢量的失真概率密度的隐马尔可夫模型(HMM)的学习过程中的失真概率密度。 (B)获得状态分割序列;参照状态分段学习观察序列,计算每个状态i的平均长度mi;状态i中每个标准化时间间隔j的平均向量υ I> ij Sub> I>。对于每个学习观察序列k,在每个状态i中,针对每个归一化时间间隔j,获取平均矢量的失真eijk;并获得最小值( n I> ij Sub> I>)和最大值( x I> ij每种状态i中每个归一化时间间隔j的总k个学习观察向量的失真。如上所述,本发明提供了比普通HMM更好的语音识别率。
展开▼