Abstract: The present invention relates to a method for recovering target speech from mixed signals, which include the target speech and noise observed in a real-world environment, based on split spectra using sound sources' locational information.
Abstract: The present invention relates to a method for recovering target speech from mixed signals, which include the target speech and noise observed in a real-world environment, based on split spectra using sound sources' locational information.