基于改进PNCC特征和两步区分性训练的录音设备识别方法

发布时间：2018-05-31 12:56

本文选题：数字音频取证 + 录音设备识别　；参考：《电子学报》2014年01期

【摘要】：录音设备来源识别是通过分析已获取的数字语音信号从而确定其录制设备的一种技术,属于数字音频盲取证.本文提出了一种基于改进PNCC特征和两步区分性训练的录音设备识别方法,由于音频中的静音包含了完整的设备信息,且不受说话人和文本等因素的影响,因此从静音段提取改进的PNCC特征,利用了PNCC的长时帧分析去除背景噪声对设备信息的影响.在模型方面,以GMM-UBM为基准模型,并通过两步区分性训练调整集内设备模型和通用背景模型,提升模型区分能力.该方法对于30种设备闭集识别的平均正确识别率为90.23%;对于15个集内和15个集外设备的测试,等错误率为15.17%,集内平均正确识别率为96.65%,验证了本文算法的有效性.
[Abstract]:Source identification of recording equipment is a kind of technology to determine the recording equipment by analyzing the acquired digital speech signal. It belongs to digital audio blind forensics. In this paper, a recording equipment recognition method based on improved PNCC features and two-step discriminative training is proposed. Because the silence in audio contains complete equipment information and is not affected by the speaker and text, etc. Therefore, the improved PNCC feature is extracted from the mute segment, and the influence of background noise on the equipment information is removed by using the long-time frame analysis of PNCC. In the aspect of model, the ability of model differentiation is improved by using GMM-UBM as the benchmark model and adjusting the device model and the general background model through two-step discriminative training. The average correct recognition rate of this method is 90.23 for the closed set recognition of 30 kinds of devices, 15.17 for 15 in-set and 15 off-set devices, and 96.65 for the average correct recognition rate in the set, which verifies the validity of the algorithm in this paper.
【作者单位】：华南理工大学电子与信息学院;卡内基梅隆大学计算机学院;
【基金】：国家自然科学基金(No.60972132,No.61101160) 广东省自然科学基金(No.9351064101000003,No.10451064101004651)
【分类号】：TN912.3

【参考文献】