Identification of Electronic Disguised Voices in the Noisy Environment

Since voice disguise has been an increasing tendency in illegal application, which brings great negative impact on establishing the authenticity of audio evidence for audio forensics, especially in noisy environment. Thus it is very important to have the capability of identifying whether a suspected voice has been disguised or not. However, few studies about identification in noisy environment have been reported. In this paper, an algorithm based on linear frequency cepstrum coefficients (LFCC) statistical moments and Formant statistical moments is proposed to identify such condition. First, LFCC statistical moments including mean values and variance unbiased estimation values, and Formant statistical moments including mean values are extracted as acoustic features, and then Support vector machine (SVM) classifiers are used to separate disguised voices from original voices. Experimental results verify the excellent performance of the proposed scheme in the noisy environment.