Dual MicrophoneWind Noise Reduction by Exploiting the Complex Coherence

Abstract:
This contribution presents methods for detecting, estimating and reducing the effect of wind noise picked up by a dual microphone communication device. To discriminate the sound induced by the wind stream the complex coherence properties of the two microphone signals are exploited. Therefore, a coherence analysis of recorded signals is given. Based on the phase variance of the complex coherence function a wind noise detection mechanism is presented. Furthermore, an approach for the enhancement of distorted speech is developed, which uses the phase variance and the magnitude of the complex cross power spectral density (PSD). This stage is realized by a noise PSD estimation and a dedicated spectral subtraction weighting. The reduction of wind noise by means of spectral weighting often leads to severe degradations to the speech signal. However, the evaluation of the proposed algorithm demonstrates a better performance compared to related approaches.