Construction of a joint peak-interval histogram using higher-order cumulant-based inverse filtering

Abstract

Conventional metrics used to quantify signals i n noise/hearing research rely primarily on time-averaged energy and spectral analyses. Such metrics, while appropriate for Gaussian-distributed waveforms, are of limited value in the more complex sound environments encountered i n industrial/military settings that have nonGaussian and nonstationary-distributed waveforms. Recent research has shown that metrics incorporating the temporal characteristics of a waveform are needed to evaluate hazardous acoustic environments for purposes of hearing conservation. The joint peak-interval histogram is a prospective candidate for use i n such an application. This paper shows that the joint peakinterval histogram can be obtained from an estimation of the temporal pattern of a complex noise waveform by using higherorder cumulant-based inverse filtering. INTRODUCTIONConventional metrics, such as the sound pressure level (SPL), and narrow or broadband (weighted) spectral energies, are used to evaluate the potential of an acoustic noise environment to produce a noise-induced hearing loss (NIHL Based upon some of these early data we formulated the working hypothesis that, for the same total energy and spectrum, a high kurtosis noise exposure is more hazardous t o hearing than a Gaussian noise exposure, and that this effect i s frequency dependent. The truth of this statement i s demonstrated in the Lei et al. [8] paper in which kurtosis (statistic) metrics in both the time and frequency domains were shown both to rank order the level of hearing trauma and t o reflect the frequency specificity of trauma. These results are a clear indication that, in addition to energy, temporal and peak variables are important determinants of hearing loss. Since the kurtosis statistics reflects the peak and temporal structure of a nonGaussian noise, an algorithm that would yield these metrics along with peak and interval histograms would be highly desirable elements of a hearing conservation noise measurement system.The fact that both temporal and spectral variables are important is not surprising since the cochlea has evolved into an exquisitely sensitive transducer of nonstationary stochastic signals typified by speech and music. Results such as outlined above have led to efforts to develop additional metrics which incorporate the temporal information inherent in the waveform of a noise. The joint peak-interval histogram, which shows the cumulative distribution of the peak amplitude reflections and timing intervals of the nonGaussian fluctuations in a noise waveform, is a candidate metric for quantifying noise exposures. This metric, in conjunction with conventional energy-based metrics, may prove to be useful i n the evaluation of a noise environment for the protection of hearing. Higher-order cumulant-based filtering can be used t o deconvolve noise waveforms to obtain the requisite amplitude and timing information for the construction of this proposed metric.Complex noises [7], which simulate the nonstationary and nonGaussian character...