基于谱减的Robust电话语音识别方法

Robust Telephone Speech Recognition Based on Spectral Subtraction

谱减技术是语音增强的有效方法，并被成功地应用到了车辆等环境下的Robust 语音识别中。文章对将谱减技术应用到电话语音识别中的加性噪声去除上进行了尝试。实验验证该方法有利于改进电话语音识别的性能，对不特定话音84个孤立词的实验，误识率下降了21．8％。

Spectral subtraction is an effective method of speech enhancement and has been successfully used for robust speech recognition in vehicle environment． In this papar the spectral subtraction is used to eliminate the additive noise for robust telephone speech recognition． Experiments show that the method is helpful to improve the performance of telephone speech recognitionand a 21．8％ word error reduction is obtained under testing on a speaker independent 84 isolated word task．