Sound source identification in time domain exploiting the clustering inverse beamforming

Acoustic imaging techniques allow the user to see what he/she is listening to. This paper aims at proposing an inverse procedure that allows for retrieving the evolution of the noise source identified in a beamforming map. Such approach overcomes the limit of frequency domain strategies, and opens up different application fields such as auralization, coherence analyses, etc. The source localization step is performed in frequency domain with the goal of accurately identifying the source coordinates. The corresponding time signals are subsequently obtained by convolving in time domain the microphones data with multiple input-multiple output impulse responses corresponding to the back-propagating functions identifying the receiver-source link. The formulation of the algorithm is presented in this paper and its main strengths and limitations are discussed. Applications are shown in simulated and real experiments.