基于PCANN/HMM混合结构的语音识别方法

A Method of Speech Recognition Based on PCANN/HMM

本文提出了一种基于ＰＣＡＮＮ／ＨＭＭ混合结构的语音识别方法，它采用相继几帧组成的特征参数 矢量作为语音识别ＨＭＭ的输入，能有效地在语音识别ＨＭＭ中引入帧间相关信息，同时为了改善多帧特征 输入ＨＭＭ的输出概率密度函数性能，在ＨＭＭ的前端增加语音参数压缩的主分量分析神经网络（ＰＣＡＮＮ）。 通过对多讲者汉语连续语音识别实验，证实了本文提出方法的有效性。

This paper presents a method of speech recognition Based on PCANN/HMM. In this method, the characteristic parameters of several successive frames are combined and treated as an input vector into HMM. It expects that segmental unit input HMM would be effective to describe the inter-frame correlation information and the PCANN is also proposed to further improve the performance for estimated values of output probability density function. By comparing them with traditional HMMs based on their speech recognition performance through the experiments of speaker-independent Chinese continuous recognition, the validity of the proposed approach could be verified.