Estimation of Signal-to-Noise Ratios in Realistic Sound Scenarios.

Abstract

Both in the design of hearing-device features, such as noise reduction, and in the evaluation measurements of such features, there is a need for more information about the sound scenarios hearing-device users encounter. The limitations of adaptive speech tests as outcome measures in the evaluation of hearing-device features have been discussed. The obvious alternative to adaptive speech testing is to test at a fixed signal-to-noise ratio (SNR). However, the question is which SNRs should be used if the test reflects real-life situations. The purpose of this study was to estimate SNRs in realistic sound environments encountered by hearing aid users. Research design was a descriptive study where recorded sound files were statistically analyzed. A total of 20 experienced, bilaterally fitted hearing aid users, all satisfied with their current hearing aids, made recordings for 3-4 days. The informants were instructed to record different situations in daily life and were told that all situations were of equal importance. The informants ranged in age from 18-81 yr old, and they had various occupations and varying hearing-loss configurations. The total duration of the recorded material for each informant was, on average, 84 min; the number of recordings was, on average, 17 per informant. The sound files were categorized based on the type of background noise and were analyzed with use of a manual noise estimation procedure. Segments of speech with noise present were cut out from the original recordings. Corresponding noise-only segments were also extracted. On the basis of power calculations for these two types of recorded segments, the SNR was estimated. Frequency-specific and overall SNRs, calculated both based on unweighted and A-weighted speech and noise levels, were derived. An estimation uncertainty measure was also developed. The range of SNRs found in the material was large. The most striking finding was that there were very few recorded situations where the SNR was negative or even close to 0 dB. For speech-in-babble noise, the average SNR was approximately 5 dB. The estimation uncertainty was generally low but became higher in highly fluctuating noises and at very low SNRs. The estimated SNRs were higher than previously reported. The results can be used in the design and evaluation of hearing-device features.