Wind noise reduction method for speech recording using multiple noise templates and observed spectrum fine structure

Wind noise affects the quality of the speech recording in the field. In this paper, we propose a wind noise reduction method based on a signal processing technique. In the method, spectrum envelopes of wind noise are estimated by using referential multiple wind noise templates and the low frequency band spectrum envelope of the observed signal. Then, we estimate the noise spectrum using the estimated spectrum envelope and the fine structure of the observed signal. Finally, we subtract this estimated noise spectrum from the observed signal. For evaluating the proposed method, we conducted subjective and objective evaluation. The experimental results showed that the proposed method achieved much better improvement than conventional spectral subtraction.