Multichannel Cross-talk Cancellation in a Call-center Scenario Using Frequency-domain Adaptive Filtering

We propose a multichannel residual cross-talk cancellation strategy to operate in environments where each microphone signal is dominated by one of several simultaneously active sound sources, like in a call-center scenario. The method relies on a two-step adaptive filtering scheme derived from an established GSC structure. The filter updates are performed in the frequency domain to allow an effective DFT-bin-wise adaptation control. This new scheme is applicable to an arbitrary number of competing sources and experiments conducted for three and seven speech sources show that it allows to reduce the amount of undesired cross-talk components by about 10dB, without noticeable distortion of the desired signal.