Improved speech intelligibility in noise with a single-microphone noise reduction technique

Abstract

Reception threshold for sentences (RTS) as measured by a modified version of the HINT test [Nilsson et al., J. Acoust. Soc. Am. 95, 1085–1099 (1984)] were collected on 26 hearing-impaired listeners fit binaurally with digital hearing aids incorporating a nine-channel spectral subtraction technique of single-microphone noise reduction. Thresholds were measured in quiet and noise presented at a zero degree azimuth with the subjects listening unaided, aided without noise reduction, and aided with noise reduction and no modification to the gain function when noise reduction was activated. Additional data included soundfield thresholds as a frequency-dependent measure of the impact of the noise reduction algorithm on sensitivity to soft sounds. RTS in quiet reveal a significant effect of noise reduction [F(3,69)=46.9, p<0.01] with noise reduction increasing RTS 3.4 dB relative to no noise reduction, from 40.4 dB (A) to 43.8 dB (A). RTS in noise reveal a significant effect of noise reduction [F(3,69)=18.7, p<0.01] with noise reduction decreasing RTS 1.8 dB relative to no noise reduction, from 0.0 dB S/N to 1.8 dB S/N. Soundfield thresholds reveal a loss of sensitivity mainly in frequencies below 1000 Hz. The noise reduction algorithm will be reviewed and discussed.