Abstract:
Elimination of observer bias is an important advantage of automating the detection of spontaneous otoacoustic emissions (SOAEs), particularly when exploring differences in SOAE prevalence across subject populations. An objective procedure was developed for detecting peaks in the power spectrum, and estimating the likelihood that such peaks arose from SOAES. The procedure involved multiple passes. First, a smoothed estimate of the spectral noise floor was obtained by replacing spectral regions containing possible SOAEs with linear estimates extrapolated from adjacent frequency regions containing no peak deviations larger than a designated criterion, and then calculating a running average on that smoothed function. Potential SOAEs were then compared with this smoothed noise floor. Based on the variance of points in the original spectrum, the likelihood that any spectral peak did not occur by chance could be estimated. The efficiency and power of this procedure was explored for several combinations of smoothing procedures and criteria for eliminating suspected peaks in the first pass. The implications of varying the criterion for acceptance of a peak as an SOAE are discussed in terms of possible differences in SOAE prevalence in population samples. [Supported by NIDCD Grant DC 00153.]