Embedding indexing information in audio signal using watermarking technique

We propose a watermarking scheme applied to digital audio signals. The goal of the project is to hide information which can identify both the tracks and the user who requests them. This technique is used for automatic identification of audio signals in broadcast programs. There are two main features that have to be satisfied: inaudibility and robustness to common editing operations. The algorithm works in a hybrid way: the watermark masking is in the frequency domain while the insertion is in the time domain. This approach allows detection to be performed directly in the time domain. In order to insert the watermark, the signal is split into frames, each of which hides binary information inserted using a spread spectrum sequence. To improve robustness, we use an error correcting code which allows the reduction of error probability. We discuss first the watermark embedding system, then the detection approach and finally the main results are presented.