基于BP神经网络的录音地点识别方法

现有的数字音频取证技术很难做到录音地点的识别,因此司法机关就不易对音频证据的有效性做出判断.针对现状,本文设计了一种基于BP神经网络的录音地点识别方法.该方法是将电网频率(ENF)作为识别根据.进行地点识别操作时,首先将电网ENF作为训练样本训练BP神经网络,然后从待取证的音频文件中提取电网频率数据并作为输入样本,用训练好的BP神经网络对输入样本进行识别,最后用模拟退火算法从识别结果中搜索出最佳识别结果,从而识别出录音的地点.实验结果表明,该方法的识别准确率最低达到90.6%,可靠性满足一定的要求.

An Approach to Identifying Rceording Locations Based on BP Neural Networks

The existing digital audio forensics technology has difficulty in identifying the location,where the recordings are made,making it hard for judicial organs to assess the effectiveness of the audio evidence.To address such an issue,this paper devises a method for identifying such locations using a grid ENF based on BP neural network. In the identification process ,grid ENF is used as a training sample for purpose of training BP neural network.Next,the grid frequency data are extracted from audio files as input samples,which are then identified by using the trained BP neural network.Finally,to identify the location of the recording,optimal recognition results are obtained from the recognition results by adopting a simulated annealing algorithm.The experimental results show that recognition rate of this approach is at least 90.6 %,and the approach is reasonably reliable.