Wind noise annoys hearing-aid users, and it is hard to attach a windscreen to a hearing-aid microphone, for cosmetic reasons. Some hearing-aid devices reduce the low-frequency components of input signals by using high-pass filters to suppress the wind noise. Although wind noise can be attenuated by this approach, the perceived binaural information of the desired signals will also be degraded simultaneously, resulting in partial information loss. We had previously proposed a short-time fast-Fourier-transform-based (STFT-based) binaural wind noise cancellation algorithm that preserves binaural cues. This algorithm required a frame length of 32 ms to maintain a high frequency resolution. However, it is known that the tolerable group delay for mild hearing loss should be less than approximately 5 ms, in the high-frequency region. In this paper, we propose a low-delay binaural wind noise cancellation algorithm that uses a frequency-warping filter. The processing latency of this algorithm is shorter than the tolerable delay. The objective evaluation results – signal-to-noise ratios and perceptual evaluation of speech quality (PESQ) scores – were improved while maintaining a low latency. Subjective experiments demonstrated that the proposed method produced almost the same score as our previous STFT-based method, in terms of the directionality of output signals.
View full abstract