A zooming-noise suppressor with no a priori information for digital still cameras

This paper proposes a zooming-noise suppressor with no a priori information for digital still cameras. Spectral peaks of the noisy speech are first detected as relevant information and preserved. Other components are suppressed to an estimated ambient-noise level. After these process of noise suppression in magnitude, the accompanying phase is randomized to make the phase-originating artifacts inaudible. Subjective evaluation results show that the proposed zooming-noise suppressor achieves a modified CCR score of 2.0.