The Research On Information Hiding Methods Based On DNN Speech Recognition Model

Communications of various multimedia such as audio and image are becoming more and more common,which inevitably brings about the problem of communication information security issues,and the information hiding technology is expected to provide technical protection for communication information security.At present,most information hiding technologies are aimed at the research using images as the carrier,while the research on audios are relatively few.Research on audio-based information hiding technology is still in the preliminary stage.As an important medium in people's daily life communication,audio provides a lot of space for information hiding,which makes the research of audio information hiding techniques valuable.Current audio information hiding techniques can be roughly classified into time domainbased and transform domain-based techniques.The time domain-based techniques usually have large hiding capacity,but the imperceptibility is not well.The transform domain-based techniques usually have better imperceptibility,but the hiding capacity is poor.To balance the hiding capacity and imperceptibility of traditional audio information hiding techniques,we propose a new audio information hiding technique based on deep neural network(DNN)-based automatic speech recognition(ASR)model.The main works of the paper are as follows:(1)Propose a new audio information hiding technique based on DNN-based automatic speech recognition model,which takes the original audio signal as input and obtains the audio signal embedded with hidden information(called stego-audio)through the training of our private DNN-based ASR model.(2)We conduct experiments to evaluate the proposed method's performance metrics of hiding capacity,imperceptibility,security,robustness and undetectability.The experimental results show that the proposed audio information hiding technique has a high hiding capacity of 48 cps with good imperceptibility,undetectability and high security.(3)We utilize the network programming to complete the design of a prototype system based on the Client-Server structure of the audio information hiding communication system,which implements user login and authentication,audio information collection and transmission,and embedding and extracting of the hidden information.(4)In the section of summary and outlook,we propose a potential application of the stegoaudio generated in this paper,which can be used to activate the intrinsic backdoor of the DNNbased ASR model,which poses a serious threat to the intelligent speakers.