In this paper, we propose a new watermark-assisted method for the analysis of audio source signals present in a mixture. This work is motivated by the issue of quality-constrained source parameters estimation in under-determined mixtures where the blind approach is not efficient. Our method uses a specific coder-decoder configuration where the separated source signals are assumed to be known at the coder. The necessary information required by a classic blind estimator to reach a target quality is imperceptibly embedded into the mixture signal. At the decoder, where the isolated source signals are unknown, the analysis process is assisted by the extra information embedded into the mixture signal. Thus, this technique aims at opening new perspectives for quality-based audio applications like active listening, sound feature extraction, or high quality audio effects with a minimal amount of side information.
