Robust wind noise detection

Wind noise can be a major problem with audio devices such as hearing aids, cochlear implants, phones and headsets. Previous wind-noise detection algorithms generally assume that large level and/or phase differences between two microphones indicate wind noise, while small differences indicate its absence. However, differences may exist without wind noise due to unmatched microphones, acoustic reflections, or the phase shift caused by the microphone spacing. This paper shows that previous algorithms do not always correctly differentiate between wind and non-wind causes of microphone signal differences, which could lead to the inappropriate engagement of wind-noise reduction processing. A novel algorithm is presented, which performs an efficient statistical analysis of the microphone signals that is substantially more robust against non-wind causes differences, and hence false wind-noise detection, in an exemplary hearing-aid application.