论文部分内容阅读
本文在统一的框架下描述了隐马尔柯夫模型(HMM)用于语音识别时的各种形式,包括离散HMM、连续混合密度HMM、半连续HMM和最大分量连续HMM等,指出各种模型均是统一形式下的导出形式.文中就离散HMM、连续混合密度HMM和最大分量连续HMM在非特定人全音节汉语语音识别中的应用,从识别率和复杂度两方面进行了性能比较.为提高最大分量连续HMM的识别性能;提出了一种修正的训练算法.
In the framework of unified framework, this paper describes various forms of Hidden Markov Models (HMM) for speech recognition, including discrete HMM, continuous mixed density HMM, semi-continuous HMM and maximum component continuous HMM. It is pointed out that all models Is a unified form of the export form. In this paper, the applications of discrete HMM, continuous mixed density HMM and maximum component continuous HMM in non-specific human syllable Chinese speech recognition are compared in terms of recognition rate and complexity. In order to improve the recognition performance of the maximum component continuous HMM, a modified training algorithm is proposed.