多带抗噪声语音识别算法研究  

A Research on Multi-band Method for Robust Speech Recognition

作　　者：孙暐[1] 吴镇扬[1] 

机构地区：[1]东南大学无线电系,南京210096

出　　处：《信号处理》2006年第4期559-563,共5页Journal of Signal Processing

基　　金：国家自然科学基金(60272044);973计划(2002CB312102)资助项目

摘　　要：根据Flether等人的研究,基于感知独立性假设的子带识别方法被用于抗噪声鲁棒语音识别。本文拓展子带方法,采用基于噪声污染假定的多带框架来减少噪声影响。论文不仅从理论上分析了噪声污染假定多带框架在识别性能上的潜在优势,而且提出了多带环境下的鲁棒语音识别算法。研究表明：多带框架不仅回避了独立感知假设要求,而且与子带方法相比,多带方法能更好的减少噪声影响,提高系统识别性能。According to the researches of Flether, etc, some algorithms based Fletcher-Allen Principle were applied to robust speech recognition. This paper replaces sub-band method with multi-band method to reduce the effect of noise. This paper theoretically analyzes the predominance of the performance of multi-band method, and presents the new muhi-band robust speech recognition algorithm. Researches show that the multi-band algorithms not only discard the perception independent assumption, but also improve recognition performance more effectively than the sub-band analysis.

关 键 词：语音识别 隐马尔可夫模型 听觉场景分析 

分 类 号：TN912.34[电子电信—通信与信息系统]