A perceptual based input decorrelation for monophonic acoustic echo cancellation

In this paper, we focus on the problem of convergence rate acceleration of adaptative algorithms in case of high correlated input signals. More precisely, we propose a new method for input signal decorrelation without any modification of the auditive quality. This method consists in adding, to the input signal, a suitable noise: it is then a solution based on the watermarking concept, where the purpose of the mark is to reduce the watermarked signal correlation. The main advantage is that no external data is used since the watermark is extracted from the masking curve. Furthermore, it is processed by a prewhitening filter in order to improve the watermarked signal correlation properties. We describe this method in some details and we show that the correlation is well reduced. The application of this method to the acoustic echo cancellation (AEC) in the mono-channel case shows a significant convergence rate improvement, when compared to that of other methods using auditory masking properties.