改进型重复Wiener滤波/PUM模型——实现抗噪连续语音识别  

Improved Iterative Wiener Filters and Posterior Union Model Robust Speech Recognition

作　　者：林劼[1] Ji Ming 刘玓[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都610054 [2]英国贝尔法斯特女王大学计算机科学系

出　　处：《计算机工程与应用》2006年第11期41-44,79,共5页Computer Engineering and Applications

摘　　要：众所周知,抗噪问题是现在语音识别研究中的重点。文章描述了一种新的抗噪语音识别方法,即通过改进型重复Wiener滤波结合后验概率联合模型PUM(PosteriorUnionModel)[3]实现在噪声环境下连续字语音识别的方法。这种方法先采用改进型重复Wiener滤波器对语音信号进行语音增强预处理,消除已知噪声,为PUM模型提供只有局部频带被噪声污染的语音信号,再利用PUM模型进行抗噪语音识别。试验表明在各种不同的噪声环境下新方法有更高的平均识别率。In this paper we present a new approach towards robust automatic speech recognition in adverse conditions. This new method is based on the combination of a speech enhancement using Improved Iterative Wiener Filters,and Posterior Union Model （PUM） to dynamically modify the probability computations performed in GMM recognizers. Previous work has demonstrated that the PUM succeeds in robust speech recognition with some interruptions and bandlimited noises. In this paper,we show that this new method more improves ratio of speech recognition in wide-band noisy conditions.

关 键 词：重复 WIENER 滤波 PUM模型 抗噪语音识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]