Using blind signal processing algorithms to remove wind noise from environmental noise assessments: A wind turbine amplitude modulation case study

Abstract

Microphone wind noise can corrupt outdoor measurements and recordings. It is a particular problem for wind turbine measurements because these cannot be carried out when the wind speed is low. Wind shields can be used, but often the sound level from the turbine is low and even the most efficient shields may not provide sufficient attenuation of the microphone wind noise. This study starts by quantifying the effect that microphone wind noise has on the accuracy of two commonly used Amplitude Modulation (AM) metrics. A wind noise simulator and synthesized wind turbine sounds based on real measurements are used. The simulations show that even relatively low wind speeds of 3 m/s can cause large errors in the AM metrics. Microphone wind noise is intermittent, and consequently, one solution is to analyze only uncorrupted parts of the recordings. This paper tests whether a single-ended wind noise detection algorithm can automatically find uncorrupted sections of the recording, and so recover the true AM metrics. Tests showed that doing this can reduce the error to ±2 dBA and ±0.5 dBA for the time and modulation-frequency domain AM metrics, respectively.