Wind Noise Reduction: The Problem, A Model, Approaches

Authors:

Nelke, C. M.

Book Title:

Colloquium on Audio Signal Processing

Organization:

Imperial College London

Date:

Jul. 2014

Note:

Invited Talk

Language:

English

Abstract

The advances in mobile communication make it possible to use communication devices (e.g. mobile phones or hearing aids) in nearly every situation. This is of course beneficial for the user, but might lead to problems concerning the quality of the conversation. A variety of acoustic scenarios such as reverberation or background noise might severely degrade the speech quality. Considering an outdoor scenario, a special class of noise is generated by a wind stream close to the microphone of the device. These noise signals clearly differ from other noise types in terms of their spectral and temporal characteristics. Thus, wind noise cannot be sufficiently reduced with general noise reduction approaches.

In this talk, both single- and multi-microphone algorithms are considered. First an analysis of the generation and the properties of wind noise signals is carried out. It will be shown that the temporal and spectral characteristics can be reproduced by a simple auto regressive model. Based on the analysis, detection mechanisms are presented, which aims to detect segments with wind noise and in the case of simultaneously occurring speech signals, estimate the degree of degradation. The last part of this talk deals with the efficient reduction of wind noise for degraded speech signal. The special characteristics of speech and wind are taken into account for the noise power spectral density estimation. Besides the standard spectral weighting, a new approach for the reduction of wind noise is presented which uses techniques known from artificial bandwidth extension.

Download

BibTeX

---