Perceptual evaluation of the functional and aesthetic degradation of speech by wind noise during recording

Abstract

This paper will present results from a systematic investigation into functional and aesthetic audio quality of speech recordings degraded by wind noise. The major source of wind noise tested comes from velocity fluctuations interacting with the transducer, generating pressure fluctuations at the microphone diaphragm. To better understand the effect of this type of noise, a perceptual experiment was designed to assess task performance and perceptions of quality when speech and simulated wind noise are presented together. A wind noise simulator was developed, which produces realistic audio from anemometer data, to allow the noise to be isolated from other ambient sounds, and also enable salient parameters to be controlled. Two key components of wind noise in recordings were evaluated, the average level and its temporal variance or “gustiness.” Eight levels of wind noise were factorially combined with three levels of gustiness. Each of these permutations was then presented with one of 24 randomly assigned, grammatically correct, nonsense sentences. Participants were asked to type the sentence they heard, rate the difficulty of the task, and indicate overall quality of the clip. Each sentence contained four keywords—correct identification of which was used for scoring performance.