摘要
文章拟解决音频取证中录音设备判别的问题。不同设备采用的压缩算法不同,会导致录音设备录音时附加到音频中的某些参量不同。从压缩算法角度,文章介绍一种基于卷积神经网络(Convolution Neural Networks,CNN)的录音设备的判别研究。首先获得不同比特率的音频,结合改进的梅尔频率倒谱系数算法(Mel Frequency Cepstrum Coefficient,Mel),用以分析录音设备对音频文件的特征参数影响,接着构建卷积神经网络识别模型,并将得到的Mel倒谱参数输入至构建好的神经网络中训练测试,最后识别并统计识别结果。实验结果表明,本研究对手机比特率的识别率达到92%。
In order to solve the problem of distinguishing audio recording equipment in audio forensics, from the perspective of the compression algorithm, this paper introduces a recording equipment identifying re- search based on CNN. Different devices have different compression algorithms leading to some personality char- acteristics attached to a recording device. First, get different audio bit rate, combined with improved MFCC to analyze the characteristic parameters of the recording device for audio files. Then build a convolution neural net- work identification model and the MFCC will be input to the neural network to test and train. The experimental results show that the recognition rate of the research to the mobile phone bit rate is 92%.
出处
《信息化研究》
2016年第2期51-54,共4页
INFORMATIZATION RESEARCH
关键词
音频取证
压缩算法
比特率
卷积神经网络
audio forensic
compression algorithm
bit rate
convolution neural network
