Energy Detection Model Research Articles

Roving-level tone-in-noise detection.

The detectability of tones, or of intensity increments to tones, in bands of random noise was measured for conditions in which the overall level was fixed or was randomly roved from interval to interval of every experimental trial. The purpose of the within-trial rove was to limit the usefulness of a detection strategy based on overall level or level within a single "critical band." At "supracritical" bandwidths, the functions relating masked threshold to noise bandwidth for the roved conditions were similar to those obtained when no rove was employed. At "subcritical" bandwidths, thresholds were higher in some roved conditions, but, for the largest rove, were still lower than would be predicted from arguments based purely on level detection--with one exception. A comparison of observer performance relative to the statistical limits imposed by the roving-level procedure indicated that the traditional critical-band energy-detector model could not account for the results, which are attributed to discrimination based on spectral shape or on waveshape.

A reexamination of the frequency discrimination of random-amplitude tones, and a test of Henning’s modified energy-detector model

A surprising result is reported by Henning [G. B. Henning, J. Acoust. Soc. Am. 39, 336–339 (1966)]: For frequencies lower than 4000 Hz, frequency discrimination was not impaired by the introduction of random differences in level between the two tones to be discriminated in a two-interval forced-choice (2IFC) task. This result is inconsistent with a model of detection and discrimination proposed by Henning [G. B. Henning, J. Acoust. Soc. Am. 42, 1325–1334 (1967)], one of a class of models in which the observer monitors the output of a single auditory filter, as well as with excitation-pattern models of frequency discrimination. In the first experiment reported here, however, an impairment of frequency discrimination with random differences in level is found when a within-subjects experimental design is used. In a second experiment, the role of pitch-intensity relationships in an experimental situation similar to that of Henning (1966) is explored. Finally, in a third experiment, an independent test of this model, which does not involve level differences, is carried out. The results are not consistent with the single-filter model, but can be accounted for by a version of the model that incorporates two filters.

Dolphin sonar target detection in noise

Several different types of experiments have been conducted to determine the sonar detection capability of the Atlantic bottlenose dolphin. In one experiment, a dolphin's performance in detecting the presence or absence of a 7.62-cm-diam (− 28.3-dB target strength), water-filled, stainless steel sphere as a function of range was determined. The threshold range in Kaneohe Bay, Oahu, Hawaii was approximately 113 m. In another experiment, the same sphere was fixed at a range of 20 m, and the detection capability of two dolphins was determined as a function of the level of artificial masking noise. In a third experiment, a simulated phantom electronic target was used to simulate a target at 20 m, and the animal's detection performance as a function of noise was determined. Using the transient form of the sonar equation, it was found that the dolphins' performances as a function of the echo signal-to-noise ratio were very similar. Dolphins process echoes like an energy detector with an integration time of approximately 264 μs. The results from the various experiments conformed to the Urkowitz [Proc. IEEE 55, 523–531 (1967)] energy detector model. The dolphin detection sensitivity was also found to be approximately 6 to 8 dB lower than that of an ideal or matched-filter receiver.

Tests of a modified energy‐detector model of frequency discrimination

Henning [J. Acoust. Soc. Am. 42, 1325–1334 (1967)] proposed a modified energy‐detector model of auditory discrimination and detection. This model, as well as several other models, predicts that random variations of the amplitudes of the signals to be discriminated in a frequency‐discrimination experiment should make the task much harder. However, Henning [J. Acoust. Soc. Am. 39, 336–339 (1966)] reported that, for signal frequencies less than 4000 Hz, variations in amplitude did not lead to the expected impairment on the frequency‐discrimination task. A reexamination of this question indicates that when the appropriate control comparisons are made, subjects do in fact show the predicted impairment. Nevertheless a second experiment suggests that this result may not provide a good test of Henning's (1967) energy detection model since variations in pitch with intensity, which are not considered in the model, are likely to be involved. For this reason, a third experiment was conducted in order to provide an independent test of the model. The results of this third experiment do not support the energy‐detector model.

The effect of masker bandwidth on signal detectability

The effect of masker bandwidth on the detectability of narrow‐ and wideband signals was investigated. The signals were pure tones and noise bands. The noise bands varied from 50–792 Hz, center frequency being 1500 Hz. Stimuli were computer generated and the noise bands had nearly rectangular spectra. Log‐log plots of threshold as a function of masker bandwidth (Fletcher plots) were found to be well fitted by two line segments. An iterative least‐squares procedure was employed to determine slopes and intercepts of the line segments. These parameters were found to depend on the bandwidth of the signal. An energy detection model [Green, J. Acoust. Soc. Am. 32, 121–131 (1960)] was modified by the addition of internal variability that is bandwidth‐dependent. This modified energy‐detection model was able to predict the results obtained with narrow bandwidth signals. The relative magnitude of internal variability was estimated and the ratio of internal‐to‐external variability agreed with estimates presented by Raab and Goldberg [J. Acoust. Soc. Am. 57, 437–447 (1975)]. In contrast, the results obtained with wideband signals were not predicted by the simple model. They may reflect the operation of a mechanism for the enhancement of spectral contours.

Detection of simple and complex tones in fixed and random conditions

Psychometric functions for detection of pure (220, 1100, or 3850 Hz) and complex tones in the presence of a 64 dB SPL uniformly masking noise were measured in a 21, 2AFC paradigm. The complex tone consisted of 18 equally intense components spaced about one critical band apart between 110 and 7260 Hz. In experiment 1, all levels for a single signal were presented in mixed order within a block of trials. Results for eight normal listeners show parallel psychometric functions for simple and complex tones. Thresholds are 43–44 dB SPL for the pure tones and about 37 dB SPL per tone for the complex tone. In experiment 2, all levels and signals were presented in random order within each block of trials. Results for four normal listeners show psychometric functions parallel to those in experiment 1. Thresholds are about 46 dB for the pure tones and 40.5 dB for the complex tone. These results are consistent with a multiband energy-detector model in which the decision is based on an unweighted sum of decision variables across an optimum selection of channels. [Supported by Deutsche Forschungsgemeinschaft and NIH-NINCDS R0 1NS 18280.]

Auditory profile analysis of irregular sound spectra.

The discrimination of changes in the shapes of sound spectra is reported. The change was always an intensity increment to the 948-Hz component of a multitone complex. First, the ability of naive listeners to learn to discriminate a change in a "regular" background or reference spectrum (equal-level tones equally spaced in logarithmic frequency) was measured as a function of the number of trials. On the average, threshold improved about 10 dB over 3000 trials, with about 50% of the decrease in threshold occurring during the first 750 trials. In a subsequent series of experiments, the overall pattern of spectral shape of the background was varied randomly. Two kinds of perturbations in spectral shape were employed: Randomly choosing the frequencies of the reference spectra and randomly choosing the amplitudes of the components of the reference spectra. The experimental manipulations involved fixing the random spectra across a block of trials, varying the reference spectra from interval to interval of each trial, and providing extensive practice in discriminating specific randomly perturbed reference spectra. The results of the spectrum-learning and random perturbation experiments provide insight into the roles of critical band filtering, sensory variability, and short-term and long-term memory representations in auditory profile analysis. Further, the appropriateness of the generalization of a simple energy detection model is discussed.

Release from masking caused by envelope fluctuations.

This paper examines how short-term energy fluctuations in a masker affect the thresholds for tones at frequencies above those of the masker. Two equally intense tones at 1060 and 1075 Hz produce up to 25 dB less masking than does a 1075-Hz tone set to the overall level of the two-tone complex. At wider frequency separations, two-tone complexes also produce less masking than the pure tone. These results indicate that envelope fluctuations in a masker, whose spectrum is confined to a single critical band, may result in release from masking. The release from masking probably is related to the comodulation masking release reported by Hall et al. [J. Acoust. Soc. Am. 76, 50-56 (1984b)] for modulated-noise maskers with bandwidths greater than one critical band. Further measurements with maskers, whose intensity level in the critical band around 1 kHz was 90 dB SPL, show similar masking by a pure tone and a 625- to 1075-Hz bandpass noise, but less masking by narrow-band noises. These results are inconsistent with a simple frequency selective energy-detector model and indicate that the auditory system can use periods of low masker energy as brief as a few ms to enhance detection of a tone. The results also imply that the upward spread of excitation is best represented by masking patterns for noises with bandwidths of several critical bands.

Binocular contrast summation—II. Quadratic summation

Quadratic summation is presented as a rule that describes binocular contrast summation. The rule asserts that for left-eye and right-eye contrasts C L and C R , there is an effective binocular contrast C given by the formula: C = ( C L ) 2 + ( C R ) 2 . Pairs of left-eye and right-eye stimuli that produce equal values of C are equivalent. Quadratic summation is applied to the results of experiments in which stimuli presented to the two eyes differ only in contrast. It provides a good, first-order account of binocular summation in contrast detection, contrast discrimination, dichoptic masking, contrast matching and reaction time studies. A binocular energy-detector model is presented as a basis for quadratic summation.

Just-noticeable differences of frequency for masked tones.

Frequency discrimination was measured as a function of level for tones of 500, 1000, 2000, 3000, and 4000 Hz. The tones were masked by a wide-band noise whose level was set to maintain constant E/No. Frequency discrimination was found to improve with level for tones of 500 and 1000 Hz while it grew worse for tones of 3000 and 4000 Hz. The data for 500 and 1000 Hz are accounted for by the periodicity-type neural timing model of Green and Luce ["Counting and Timing Mechanisms in Auditory Discrimination and Reaction Time," in Contemporary Developments in Mathematical Psychology, edited by D. H. Krantz, R. C. Atkinson, R. D. Luce, and P. Suppres (Freeman, San Francisco, 1974). Vol. 2]. The data for 3000 and 4000 Hz are explained by the modified energy detection model of Henning [J. Acoust. Soc. Am. 42, 1325-1334 (1967b)].

Temporal versus spectral cues in pitch perception of AM noise

The phenomenon that periodically interrupted white noise exhibits a weak pitch corresponding to the interruption rate has traditionally been regarded as evidence for time‐domain pitch processing by the auditory system. Recently, however, it has been pointed out [Pierce, Lipes, and Cheetham, J. Acoust. Soc. Am. 61, 1609–1621 (1977)] that the short‐term spectrum of such a stimulus contains several kinds of potential pitch cues. Melodic interval identification experiments were performed with sounds consisting of low‐pass filtered noise. modulated by a periodic signal. The modulation signal was either a sine wave, square wave, or pulse train. Melodic intervals were generated by varying the frequency fm of the modulation signal. Modulation frequencies were in the 100–400 Hz range. The low‐pass cutoff frequency of the noise, fco, was used as an experiment variable. Identification performance shows a particular dependence on the ratio fco/fm for each type of modulation signal. The form of this dependence is shown to be consistent with both a modified energy detector model (time domain) and a short‐term power spectrum correlation model (frequency domain). [Work supported by NIH.]

Amplitude-modulated noise: the detection of modulation versus the detection of modulation rate.

Modulation threshold, that is, the modulation depth required to discriminate a sample of amplitude-modulated (AM) noise from a sample of unmodulated noise, was measured as a function of modulation rate (16--320 Hz), modulator waveform (sine or square), and the bandwidth of the AM noise (0.5--8.0 kHz). Modulation threshold increases monotonically with modulation rate, sine-wave thresholds are greater than square-wave thresholds, and threshold rises as the bandwith of the AM stimulus decreases. These effects all support the use of some form of energy detection model to explain modulation threshold. The modulation thresholds were compared with pitch thresholds gathered under precisely the same conditions. Pitch threshold or, alternatively, rate threshold was taken to be the modulation depth required to decide which of two samples had the higher modulation; the rate difference was 20%--just over three semitones. In the region above about 70 Hz, rate threshold is essentially a constant multiple of modulation threshold, indicating that the primary constraint on rate threshold is the audibility of the modulation. Below 70 Hz, rate and modulation threshold diverge; it is argued that the limit on rate threshold in this region is probably the length of the correlation required to extract the periodicity.

Frequency discrimination for tones of different levels but constant E/N0

Wever's theory of pitch perception asserts that both place and periodicity mechanisms are used by the auditory nervous system, with periodicity predominant at low frequencies and place predominant at high. One way of gaining insight into the coexistence of the two mechanisms is to find an experimental parameter that, when varied, produces differential effects for low frequencies and high. To this end, signal level was varied and frequency discriminability was measured for tones of 500, 1000, and 4000 Hz. In order to examine the effects of level independent of E/N 0, the level of a noise masker was chosen so that the signal‐to‐noise ratio remained constant. Raising level was found to improve frequency resolution in the 500and 1000‐Hz conditions, while performance became worse in the 4000‐Hz condition. The high frequency data can be explained in terms of Henning's modified energy detector model, with the added assumption that filter width increases with level. The low‐frequency data are accounted for by a level‐dependent increase in the number of active channels in the counting model of Luce and Green.

Stimulus variability and auditory filter shape.

We have investigated the way in which stimulus variability will affects the attenuation characteristic or auditory filter shape inferred from masking experiments. Stimulus variability was found to have a pronounced effect on filter shapes derived from bandlimiting experiments in which the signal is a tone, the masker is a band of noise centered on the tone, and the independent variable is the width of the noise band. But stimulus variability had virtually no effect on filters obtained from notched noise masking experiments. In an attempt to extend the utility of the filter-shape concept we have replicated an earlier experiment in which the masker is two tones rather than noise. The tones, each 57 dB SPL, were used to mask a narrow band of noise centered midway between them and threshold for the noise signal was measured as a function of the frequency separation of the tonal maskers. The form of the data is in good agreement with the auditory filter shape derived using a notched-noise masker and a tonal signal. In order to predict the absolute levels obtained in the noise-masking experiments, it was necessary to assume that internal noise adds to the variability of the stimuli. The resulting model, which incorporates the filter shape from the notched noise experiment and the energy detection model, predicts not only the form but also the absolute levels obtained in the bandlimiting and notched noise experiments. In addition, it predicts the shape of the data from the two-tone masking experiment but it does not predict the overall level.

Detection threshold for a two-tone complex

The absolute threshold for two-component tone bursts of 500-msec and 1-sec duration was measured as a function of the frequency spacing between the components. The average frequency of the two components was 700 Hz. When interpreted in terms of an energy-detector model (peripheral filter followed by squarer, integrator, and threshold detector), the measurements give an estimate of the time constant and window shape of the integrator. Our measurements on two subjects give a time constant on the order of 200 msec. A window which initially decreases more rapidly than exponentially fits the data better than an exponential window.

Detection threshold for a two-tone complex

This paper reports on a measurement of the absolute threshold for a two-tone signal as a function of the frequency separation between the tones. When interpreted in terms of an energy detection model (peripheral filter followed by squarer, integrator, and threshold detector), the measurements give an estimate of the time constant of the integrator. This time constant has been estimated by many investigators, and published estimates range from 10 to 500 msec. Our measurements on two subjects give a value in the neighborhood of 150 msec.

Thresholds for click pairs masked by band-stop noise.

The auditory threshold was measured for a pair of clicks in the presence of band-rejected masking noise. This threshold depends on the interval between the two clicks. As the interval is changed, the threshold intensity alternately increases and decreases. For homopolar clicks and masking noise with a spectral gap centered at f Hz, maxima are at (2n +1)/(2f) sec (n = 0,1,2,…). Experimental determinations of this fluctuation, for equal-intensity clicks, are in quantitative agreement with predictions of a simple model. In the model, band-stop noise produces a band pass filtering of the click stimulus, and the stimulus is detected if the long-time average energy at the filter output exceeds some threshold value. However, if the clicks are of unequal intensity, the threshold is influenced by the click order. This order effect is inconsistent with the energy-detection model.

Auditory critical bandwidth for short-duration signals.

A new experimental technique is proposed to estimate the bandwidth of the auditory-filter mechanism when detecting short-duration sinusoidal signals masked by wide-band noise. The experiment requires the use of only wide-band noise and two signals of the same frequency, power, and duration, but having different energy density spectra. The special feature of this technique is that the subjective impression made by the masking noise remains the same throughout the experiment unlike experiments that employ progressively changing bandwidths of the masking noise. This avoids the possibility of any change in the parameters of the detection mechanism due to changes in the subjective impression made by the noise. The analysis makes use of the energy detection model consisting of a bandpass filter, square-law rectifier, and an integrator to describe the auditory detection system. A rectangular filter and an LCR single tuned filter have been considered in the analysis. From the experimental results it is found that the rectangular filter provides a better description of the auditory-filter mechanism than the single tuned filter. The bandwidth estimates show an increase with signal frequency. The results also support the idea of a minimum auditory-filter bandwidth for a continuous or long-duration signal and increased values for short-duration signals depending on the signal duration.

Negative MLD's with Gated Sinewave Masking

The detectability of S0 and Sπ signals, masked by a tone of the same frequency and duration, was measured as a function of the signal-masker phase. The signal and masker frequency was 250 Hz, and the duration 1/10 sec. In all conditions, Sπ signals were less detectable than S0 signals (a negative MLD), and this effect was largest (about 10 dB) with signal and musket in phase. Detection of S0 signals as a function of signal-masker phase closely follows the predictions of a simple energy detection model. The results with Sπ signals, however, are not consistent with any existing theory of binaural detection. In some general respects, the data suggest a mechanism similar to Durlack's equalization-cancellation mechanism. [Research supported in part by the National Institutes of Health, Public Health Service, U. S. Department of Health, Education and Welfare.]