论文部分内容阅读
为了解决分段概率模型 (SPM)因缺少对时间信息描述而带来的建模精度低的问题 ,提出了状态驻留分段概率模型 (SDSPM)。SDSPM中包含了用伽玛分布表示的状态驻留概率 ,以刻划语音的时间特征。此驻留概率相当于隐马尔可夫模型 (HMM)中的状态转移概率 ,但使 SDSPM描述语音时间特征的能力强于 HMM。SDSPM既改善了 SPM的模型性能 ,同时又避免了 HMM的计算复杂度问题。测试实验证明了 SDSPM模型在汉语语音识别中的有效性。
In order to solve the problem of low accuracy of modeling the segmentation probability model (SPM) due to the lack of time information description, a state-resident segmentation probability model (SDSPM) is proposed. SDSPM contains the state-resident probability in terms of gamma distribution to characterize the temporal characteristics of speech. This resident probability is equivalent to the state transition probability in HMM, but it makes SDSPM stronger than HMM in describing the temporal features of speech. SDSPM not only improves the model performance of SPM, but also avoids the computational complexity of HMM. The test experiment proves the validity of SDSPM model in Chinese speech recognition.