A wind-noise suppressor with wind-burst detection based on a stationary noise estimate

This paper presents a wind-noise suppressor with wind-burst detection based on a stationary noise estimate. Input that exceeds an estimated stationary noise level is detected as wind burst. Upon detection, the spectral gain is reduced to apply maximum suppression determined by a floor gain. False detection of a speech onset as wind is minimized by speech-wind discrimination with spectral smoothness. Its implementation as an Android app is outlined. Subjective evaluation results show that a 5-grade ACR (absolute category rating) MOS for the proposed wind-noise suppressor reaches 3.4 and is 0.56 higher than that by a conventional noise suppressor with a statistically significant difference.