A multirate approach to multichannel blind deconvolution

Multichannel blind deconvolution techniques can be used to de-reverberate and separate audio signals recorded by arrays of microphones in multimedia applications. Experimental results are available only for simple cases, using synthetic signals or considering short and stationary impulse responses. The problem of multichannel blind deconvolution of real-world audio signals is considered, using true speech signals recorded in a real furnished room. The obtained results are very good in terms of both separation and de-reverberation, demonstrating that the problem can be effectively challenged in practice. Since the complexity of the system can easily exceed the single DSP processor capabilities, a new sub-band architecture is proposed, which shows some structural and practical advantages, especially when processor arrays are employed.