Effects of low-cut filtering on speech recognition with noise

Abstract

Many types of hearing aids apply low-cut filtering for noise reduction. This method is based on the fact that background noises are mainly low frequency. The advantages of low-cut filtering for speech recognition with noise have been contentious. We evaluated the effectiveness of the low-cut filtering method for improving the recognition of Japanese sentences with noise. Speech recognition tests were undertaken by 56 patients with sensorineural hearing loss and 14 volunteers with normal hearing. Multitalker noise was added to sounds of four Japanese sentences by a male announcer at an S/N ratio of 6 dB. Two of the four sentences were processed through a low-cut filter (1000 Hz cutoff frequency, - 12 dB per octave filter slope). In 48 of the patients with sensorineural hearing loss, the two filtered sentences were not understood better than the unfiltered sentences. In 34 of these patients, speech recognition deteriorated as a result of low-cut filtering. These results indicated that low-cut filtering is not effective for speech recognition with noise.