采用线性预测模型的语音篡改检测

基于线性预测模型,提出一种通用的语音信号真实性和完整性的鉴别方法.将线性预测残差信号通过带通滤波器,消除谐波信号分量的干扰.滤波后的原始语音残差信号呈高斯分布,而篡改语音的残差则体现出明显的超高斯特性,将预测残差的高阶统计特征作为判断篡改的依据.实验结果表明:该方法能够有效实现语音篡改盲检测,并定位篡改位置;在噪声环境下,与现有方法相比,文中的检测方法具有更高的鲁棒性.

Speech Forgeries Detection with Linear Prediction Model

Based on linear prediction model, a general forensic approach aiming to ensure the authenticity and integrity of speech is presented. To eliminate the influence of harmonic components existed in the LPC residual, a band-pass filter is introduced. The original speech residual follows a Gaussian distribution while its forgery counterpart shows a super Gaussian characteristic. Higher order statistics of the LPC residual is applied to forgery detection. Experimental results show the effectiveness of our method in detecting and locating forgery. Results also demonstrate the higher robustness of our detection method in noise environment compared to the state-of-the-art method.