【摘要】 本文探讨经短时信号处理后的语音信号帧间相关信息对基于隐马尔可夫模型（ＨＭＭ）的语音识别系统识别精度的影响．鉴于ＨＭＭ的输出独立假设导致语音帧间相关信息的损失，本文提出一种描述帧间相关信息的统计模型一马尔可夫链（ＭＣＭ），用来弥补ＨＭＭ在这方面的缺陷．经非特定人的和多话者孤立字实验表明，用ＭＣＭ作为ＨＭＭ的辅助模型，可将原有ＨＭＭ系统的识别率提高约１～６个百分点．更多还原

【Abstract】 The paper investigates the effect of interframe spectral dependence in speech on the recognition accuracy of HMM-based speech recognition systems. It is proposed that the Markov chain model (MCM)may be used to account for the spectral dependence between frames.Experiments on isolated word recognition have shown that using the MCM as a complement to the HMM in recognition can achieve improved recognition approximately by 1~6 percentage points as compared to HAM-based systems.更多还原