论文部分内容阅读
为适应语音识别的需要，作者克服了传统隐马尔可夫模型（ＨＭＭ）只考虑当前观测符号之前状态的缺点，吸收其采用“隐含”层的处理方式，将其纳入马尔可夫随机场（ＭＲＦ）的框架，建立了一个基于ＭＲＦ的语音识别模型，并较详细地阐明了这个系统的训练和识别算法，重新定义了松弛标注算法中相应的支持函数。典型实验表明，ＭＲＦ模型较传统的ＨＭＭ 有较高的识别率。在优化初始参数的条件下，两种模型的识别在同样的时间范围内。在训练脱机的情况下，ＭＲＦ模型有其明显的优势。
In order to meet the needs of speech recognition, the author overcomes the shortcomings of traditional hidden Markov models (HMMs) that only consider the state before the current observing symbol, absorbs the processing methods of “implied” layers and integrates them into Markov random fields MRF) framework, an MRF-based speech recognition model is established. The training and recognition algorithms of this system are elaborated in detail, and the corresponding support functions in the relaxation labeling algorithm are redefined. Typical experiments show that the MRF model has a higher recognition rate than the traditional HMM. Under the condition of optimizing the initial parameters, the recognition of the two models is within the same time range. The MRF model has its obvious advantages when training offline.