Sub-Band Detector for Wind-Induced Noise

Abstract

This study considers the problem of reliably detecting wind in portable devices such as mobile phones, personal digital assistants (PDA), tablet computers, and alike. A multi-microphone wind detector is presented that utilizes statistical analysis of microphone signals. This statistical analysis is used to quantify the discrepancy in dynamic behavior of the microphone signals. Empirical Distribution Function (EDF) is proposed to capture the dynamic behavior of each microphone signal, and the discrepancy in this behavior is quantified using mean absolute difference between the corresponding EDFs. A decision about wind presence is made separately for each chosen sub-band in order to avoid performing wind noise reduction in a part of audio spectrum which was not affected by wind-induced noise. The proposed wind detector demonstrates reliable detection of wind for each chosen sub-band in an exemplary mobile phone application. Performance evaluation confirms superiority of the proposed sub-band wind detector in terms of correct detection and false alarm compared to a state-of-the art wind detector.