This paper presents a hybrid framework of feature extraction and hidden Markov modeling (HMM) for two-dimensional pattern recognition. Importantly, we explore a new discriminative training criterion to assure model compactness and discriminability. T

通过基于HMM的语音识别系统的基本结构,详细介绍了语音信号采集、预处理、MFCC特征参数提取、HMM训练和HMM识别等主要模块的基本原理 并针对实际应用中HMM存在的模型初始化和数据溢出等问题进行分析,引入了一些能提高系统性能的相应措施。-By the basic structure of HMM-based speech recognition system, details the basic principles of audio signal acquisition, preproces