基于混合分类器的AMR-WB语音隐写分析方法

AMR-WB Steganalysis based on Hybrid Classifier

基于自适应多速率宽带语音编码的信息隐藏方法具有较强的隐蔽性和抗检测能力。为检测导谱频率矢量索引搜索中的量化索引调制,提出了一种基于校准技术和混合分类器的隐写分析方法。其中,一种基于长短期记忆模型的特征提取方法被用于提取导谱频率的关联性特征,导谱频率量化索引的概率分布则作为导谱频率的长时分布特征。以上两种特征通过校准技术处理后,被送入支持向量机中进行训练,可获得具有较高分类性能的分类器。实验结果表明,在不同嵌入率和语音长度下,所提方法均有较高的分类正确率。

The information hiding method based on adaptive multi-rate wideband speech coding has strongconcealment and anti-detection ability. In order to detect the quantization index modulation in the frequencyvector index search of the pilot spectrum, a steganalysis method base don calibration technology and hybridclassifier is propose. Among them, a feature extraction method based on the long and short-term memory modelis used to extract the correlation feature of the pilot spectrum frequency, and the probability distribution of thepilot spectrum frequency quantization index is used as the long-term distribution feature of the pilot frequency.After the above two features are processed by calibration technology, they are sent to the support vector machine fortraining, and a classifier with higher classification performance can be obtained. Experimental results show that theproposed method has a higher classification accuracy rate under different embedding rates and speech lengths.