一种语音信息检测方法的研究

The Research for a Speech Detection Method

文章提出了一种新的语音信息检测的较灵活的方法。其中用到了两种技术:人工神经网络和复倒谱矩阵。目的是如果用人工神经网络就能够较彻底地解决未明确定义的映射关系。对各种在较低的噪音信噪比值情况下观察结果都有较高的可信度。在语音信号检测过程中,由于语音的特征文章利用线性预测系数得到复倒谱矩阵,这样做会以最低的代价提供较高的对数频谱的估计程度,并且提高了频谱域和时域的有效性。文章测试了几种不同的W SS噪声以及不同信噪比(SNR)的情形,在3dB～10dB的范围之内,AN N方法显著地优于利用语音信号的能量和过零率检测的方法,同时也提高了其它基于复倒谱矩阵方法的准确率。

In this article a new flexible, speech detection method comprising two relatively modem approaches like Artificial Neural Networks （ANN） and cepstral matrices is presented.The purpose of this reseach is that problems that are not well defined and mathematically vague are known to be successfully solved just by the ANN approach.This method seems to work quite reliably even at low SNR and in various sorts of WSS noises.Cepstral matrices obtained via linear prediction coefficients are chosen as the eligible speech features,This technique is known to provide reliable log spectrum estimation at a low cost,Furthermore,both spectral and time characteristics can be efficient,which is an essential aim here,Several WSS noises and different SNR settings are tested.in the range of 3dB to 10dB the ANN approach remarkably outperforms the energy and zero crossing method and improves the accuracy of the other algorithm based on cepstral matrices as well.