This paper proposed smart noise reduction by understanding acoustic scene. Noise reduction is incorporated into DOA estimation, and updates a noise model, which aims at improving noise robustness in DOA estimation. Based on acoustic and spatial information extracted from multiple observations, either spectral subtraction or beamforming is carried out based on both a reliability of each DOA estimate and target signal detection. As a result of performance evaluation, the proposed adaptive noise method has a significant advantage in DOA estimation under non-stationary noise conditions over a static noise model.
展开▼