A Watermarking-Based Method for Informed Source Separation of Audio Signals With a Single Sensor

Abstract

Abstract-In this paper, the issue of audio source separation from a single channel is addressed, i.e. the estimation of several source signals from a single observation of their mixture. This challenging problem is tackled with a specific two levels coderdecoder configuration. At the coder, source signals are assumed to be available before the mix is processed. Each source signal is characterized by a set of parameters that provide additional information useful for separation. We propose an original method using a watermarking technique to imperceptibly embed this information about the source signals into the mix signal. At the decoder, the watermark is extracted from the mix signal to enable an end-user who has no access to the original sources to separate these signals from their mixture. Hence, we call this separation Index Terms-under-determined source separation, watermarking, audio processing, speech processing.

Perceptually controlled doping for audio source separation

“…For example, in [15], the time-frequency plane is divided in 'molecules' and the watermark information is either the energy contribution of each source to each molecule of the mixture or a coarse description of each molecule of each source. This watermark helps the separation of a linear instantaneous monophonic mixture of four or five sources.…”

Informed source separation: Source coding meets source separation

“…Finally, it is also quite clear now that source separation performances strongly depend on the amount of available prior information about the sources and the mixing process one can introduce in the source separation algorithm [2]. Motivated by this observation a new setting called informed source separation (ISS) [3,4,5,6] was recently considered, where both the sources and the mixtures are assumed known during a so-called encoding stage. This knowledge enables the computation of any kind of sideinformation that should be small and should help the source separation at the so-called decoding stage, where the sources are no longer assumed to be known.…”

“…Several approaches were proposed for the ISS problem [3,4,5], and a common point of these methods is that they all rely on some source model θ transmitted as a side-information. Assuming the sources to be sparse in a given time-frequency (TF) representation, Parvaix et al [4] construct a model θ that for each TF point includes the indices of the sources supposed active in this TF point.…”

“…Assuming the sources to be sparse in a given time-frequency (TF) representation, Parvaix et al [4] construct a model θ that for each TF point includes the indices of the sources supposed active in this TF point. A TF molecular dictionary is used as model θ in [3]. Liutkus et al [5] go beyond the sparsity assumption, that is hardly verified for real-world mixtures, and rather consider probabilistic models θ such as local Gaussian models (LGM) [1,2] with structured or free variances.…”

“…The best (the minimal) achievable distortion produced by conventional ISS methods [4,5] is incompressible, i.e., it is bounded below. This remark does not concern [3], where the distortion can be always decreased by increasing the size of the corresponding molecular dictionary, which would lead, however, to an excessive rate needed to transmit such a dictionary. Figure 1 gives a simplified interpretation of several model-based methods applied to a mixture of two sources in one TF point, TF indices being omitted (see figure's legend for details about notations).…”