SSLSS: Semi-Supervised Learning-based Steganalysis Scheme for Instant Voice Communication Network

Shanshan Tu, Huaizhou Tao, Yongfeng Huang, Zhili Zhou

Abstract


In the field of instant voice communication steganalysis, the traditional detecting methods are mainly based on supervised learning scheme which result in a large amount of complex manual pre-processing, and its accuracy can be easily destroyed by the difference between the distribution of training set and testing set in the actual voice application. View of the above problem, this paper firstly introduced a novel semi-supervised hybrid learning detection model for the instant voice communication network, in which the progress of manually annotating training data set has been removed to solve the problem of complex operations and poor applicability in classifier, therefore this model has a simpler structure and more extensive detection scopes. Then we designed a Multi-Criteria Fusion module, which can automatically generate the pseudo-label set from testing data set to train the classifier model, thus our scheme will not be affected by the distribution shift. In this module, we defined the confidence level and representative level to judge if a feature vector is appropriate to be pseudo-labeled. By doing experiments in low bit-rate speech coding steganalysis (G.723.1/G.729/iLBC speech codecs) on Quantization index Modulation (QIM), which are common codecs in instant voice communication network, the results showed that our method has higher accuracy than un-supervised methods, and when the distribution of training and testing data sets are different, the proposed approach is less affected than previous supervised methods and has obvious advantages in accuracy. The experiments also proved that our method can be deployed on different kinds of Instant Voice Communication (IVC) codec.


Full Text:

PDF

Refbacks

  • There are currently no refbacks.