Direction of Arrival Estimation through Noise Supression: A Novel Approach using GSC Beamforming and Room Acoustic Simulation

A novel method for localizing or estimating the direction of a sound source from the speech mixed with different levels of noise recorded by a microphone array embedded in an Unmanned Aerial Vehicle (UAV) has been proposed in this paper. Publicly available DREGON dataset (The IEEE Signal Processing CUP 2019 dataset for static task) has been used. The detail methodology of the system for localizing the sound source in static condition of the UAV is described in this paper. Generalized Sidelobe Canceller (GSC) Beamformer on the noisy audio is used to extract the noise along the rotor directions. This extracted noise is the simulated to synthesize 8 channel audio using pyroomacoustics. Finally the extracted noise is used as the reference of the wiener filter for filtering the noise in the provided noisy audios. GCC PHAT and GCC NON LIN methods are used to estimate the elevation and azimuth of the sound source. Promising results have been found using this method to localize the sound source of human speech from the audios of snr as low as −20 dB recoreded by a microphone array embedded on a UAV. If at most 10° of error in angle is allowed, our proposed method provides an accuracy of almost 91.67%.