Abstract:

To insert additional data into video and audio material, numerous methods are used based on appropriate manipulation of the spectrum, such as a wavelet transform or bit modification such as least significant bit (LSB) modification. These methods are effective, but have their limitations in the amount of data that can be added without significantly affecting the original video or audio material.

The article presented here describes the possibility of using Deep Neural Networks (DNNs) to transparently encode an audio signal into another audio signal. This paper will present the neural network structure used for this purpose, an example learning set and the results.