Abstracts Submitted: 257
Number of Users: 304
View Abstracts Submitted
Back to home Page
The number of active hearing aid users has increased, and hearing aids have been used even outdoors. According to a survey by Mark Trak Ⅷ, in windy conditions, the levels of their satisfaction stay low since wind noise degrades speech intelligibility, masking speech signals. Although general microphones can be attached windscreen for suppressing wind noise, it is hard to attach it to hearing aids for cosmetic reasons. Therefore, for hearing aids, wind noise should be suppressed by signal processing. An algorithm using the characteristics of head-related transfer functions have been proposed by authors to resolve these problems. The algorithm can effectively suppress wind noise, reducing distortion of speech signals and preserving binaural cues. However, it required frame processing whose length is 32 ms so as to maintain high resolution in frequency. The delay caused by the frame processing becomes a factor degrading the speech intelligibility because the processed sound which has 32 ms delay is heard as an echo of direct sound in the case of using like an open-fitting style hearing aid. Generally, it is known that the tolerable group delay for mild hearing loss should be below about 5 ms. This issue was resolved by using frequency warping filter-bank with analysis length of 31 points as a synthesis and analysis stage of our system. The frequency resolution of low-frequency part was kept an efficient resolution by adjusting warping parameter in order to achieve that of FFT based system with 32 ms frame length. In this report, the binaural wind noise cancellation system is implemented on a prototype DSP board to evaluate real time performance of the system in an actual environment. In the system, the number of all-pass filter for frequency warping is set to 31. The frame length of FFT is almost equivalent to its number and frame shift is 16. For objective evaluations, waveforms between input and output signals are compared and its SNR improvements are also calculated. It is revealed that wind noise is suppressed effectively by the system even under the actual environment and the SNR of the output signal is improved compared with the SNR of the input signal.
© Copyright 2017 All Rights Reserved