PSD estimation in beamspace for source separation in a diffuse noise field

An extension of the method for estimating the power spectral densities of both coherent signals and incoherent noise is proposed. The previous method utilised the different combination of the directivity gain, which is realised by applying several beamformings to the observation of a microphone array, in order to estimate the power spectral density of coherent signals. The extension introduces the diffuse sound field model in order to estimate the power spectral density of the incoherent noise as well as that of the coherent signals. Experiments for source separation and noise reduction prove the proposed method can accurately estimate the power spectral densities of both coherent signals and incoherent noise.