时频参数和模糊分类器在词边界检测中的应用

Word Boundary Detection in Variable Noise-Level Environment Based on Fuzzy Classifier

词边界检测误差是语音识别中产生错误的主要原因之一。常规的检测算法在低信噪比尤其在背景噪声能量可变的环境下不能有效工作。本文提出用语音信号的精确时频参数和过零率来训练模糊神经分类器,进行词边界检测。不同背景噪声下的实验结果表明,该方法可适应背景噪声能量的变化,得到高正确率的词边界检测。

A major cause of errors in speech recognition is the inaccurate detection of word boundary. Conventional methods cannot work well in the condition of low SNR or at a variable background noise level. This paper proposes to use refined timefrequency parameter and zero crossing rate of speech signal to train neurofuzzy classifier to detect word boundary. The experiments in different noise background levels show that the algorithm can be adaptive to the variation of the background noise level and obtains high accuracy rates.