Effects Of Signal Duration Research Articles

The Effect of Signal Duration on the Classification of Heart Sounds: A Deep Learning Approach

Deep learning techniques are the future trend for designing heart sound classification methods, making conventional heart sound segmentation dispensable. However, despite using fixed signal duration for training, no study has assessed its effect on the final performance in detail. Therefore, this study aims at analysing the duration effect on the commonly used deep learning methods to provide insight for future studies in data processing, classifier, and feature selection. The results of this study revealed that (1) very short heart sound signal duration (1 s) weakens the performance of Recurrent Neural Networks (RNNs), whereas no apparent decrease in the tested Convolutional Neural Network (CNN) model was found. (2) RNN outperformed CNN using Mel-frequency cepstrum coefficients (MFCCs) as features. There was no difference between RNN models (LSTM, BiLSTM, GRU, or BiGRU). (3) Adding dynamic information (∆ and ∆²MFCCs) of the heart sound as a feature did not improve the RNNs’ performance, and the improvement on CNN was also minimal (≤2.5% in MAcc). The findings provided a theoretical basis for further heart sound classification using deep learning techniques when selecting the input length.

Signal-to-noise ratio of diverging waves in multiscattering media: Effects of signal duration and divergence angle.

In this paper, SNR maximization in coded diverging waves is studied, and experimental verification of the results is presented. Complementary Golay sequences and binary phase shift keying modulation are used to code the transmitted signal. The SNR in speckle and pin targets is maximized with respect to chip signal length. The maximum SNR is obtained in diverging wave transmission when the chip signal is as short a duration as the array permits. We determined the optimum diverging wave profile to confine the transmitted ultrasound energy in the imaging sector. The optimized profile also contributes to the SNR maximization. The SNR performances of the optimized coded diverging wave and conventional single-focused phased array imaging are compared on a single frame basis. The SNR of the optimized coded diverging wave is higher than that of the conventional single-focused phased array imaging at all depths and regions.

Audio content analysis for unobtrusive event detection in smart homes

Environmental sound signals are multi-source, heterogeneous, and varying in time. Many systems have been proposed to process such signals for event detection in ambient assisted living applications. Typically, these systems use feature extraction, selection, and classification. However, despite major advances, several important questions remain unanswered, especially in real-world settings. This paper contributes to the body of knowledge in the field by addressing the following problems for ambient sounds recorded in various real-world kitchen environments: (1) which features and which classifiers are most suitable in the presence of background noise? (2) what is the effect of signal duration on recognition accuracy? (3) how do the signal-to-noise-ratio and the distance between the microphone and the audio source affect the recognition accuracy in an environment in which the system was not trained? We show that for systems that use traditional classifiers, it is beneficial to combine gammatone frequency cepstral coefficients and discrete wavelet transform coefficients and to use a gradient boosting classifier. For systems based on deep learning, we consider 1D and 2D Convolutional Neural Networks (CNN) using mel-spectrogram energies and mel-spectrograms images as inputs, respectively, and show that the 2D CNN outperforms the 1D CNN. We obtained competitive classification results for two such systems. The first one, which uses a gradient boosting classifier, achieved an F1-Score of 90.2% and a recognition accuracy of 91.7%. The second one, which uses a 2D CNN with mel-spectrogram images, achieved an F1-Score of 92.7% and a recognition accuracy of 96%.

Behavioral measures of cochlear gain reduction and gain reduction in with normal hearing or minimal cochlear hearing loss

This is a continuation of a study examining the relationship between cochlear hearing loss and various psychoacoustic measures thought to be related to cochlear function. In the listeners tested, thresholds for long tones ranged from well within the clinically normal range to just outside this range. Where thresholds were elevated, other clinical tests were consistent with a cochlear origin. Because the medial olivocochlear reflex (MOCR) decreases the gain of the cochlear active process in response to sound, when possible, measures were made with short stimuli. Signal frequencies were 2, 4, and 8 kHz. Maximum gain was estimated by measuring the threshold masker level for a masker at the signal frequency and a masker well below the signal frequency. The effect of signal duration on threshold was measured using 10 and 200-signals. One point on the lower leg of the input/output function was measured by finding the threshold masker level for a masker slightly less than one octave below the signal frequency needed to mask a signal at 5 dB SL. Gain reduction was estimated by presenting a pink noise precursor before the signal and masker, and measuring the change in signal threshold as a function of precursor level. The relationship between these measures will be discussed. [Work supported by NIH(NIDCD)R01 DC008327 (EAS), and grants from the Purdue Office of the Executive Vice President for Research and the Purdue Graduate School (WS).]

Duration and transition effects in multiple-burst, multitone masking

Previous studies of tone detection using a multiple-burst same (MBS) or multiple-burst different (MBD) masker have shown that tone detection thresholds were lower with an MBD masker than with an MBS masker, and decreased as the number of bursts increased. In those studies, as the number of bursts increased, the total signal duration increased as well as the number of times the (MBD) masker changed frequencies. The present study was designed to differentiate between the effects of signal duration and masker transitions. In one condition, the burst duration was inversely varied with the number of bursts so that the overall stimulus duration was always 480 ms. In a second condition, the stimuli always had eight 60-ms bursts, but the number of times the masker transitioned from one random draw of frequencies to a different draw was varied from zero (MBS) to seven (MBD). The MBD condition showed a steady improvement with number of bursts, regardless of the burst duration. There was a similar improvement with the number of masker transitions. There was only a small improvement in thresholds with increasing burst duration. Results suggest that the number of changes of masker frequencies drives performance more than the total signal duration.

High-frequency hearing in seals and sea lions and the implications for detection of ultrasonic coded transmitters

In order to better understand the ability of pinnipeds to detect acoustic signals from ultrasonic coded transmitters (UCTs) commonly used in fisheries research, high-frequency hearing thresholds were obtained from a trained Pacific harbor seal (Phoca vitulina) and a trained California sea lion (Zalophus californianus). Using a 69 kHz, 500 ms, narrow-band FM sweep stimulus, detection thresholds for the harbor seal and the sea lion were determined to be 106 dB and 112 dB re 1 μPa respectively. While the harbor seal threshold falls within the range of existing data, the sea lion threshold is 33 dB lower than expected based on previous reports. This finding indicates that sea lions may be more sensitive to the output of UCTs than previously thought, and allows for the possibility that acoustically tagged fish may be selectively targeted for predation by sea lions as well as seals. These hearing thresholds, combined with ongoing work on the effect of signal duration on high-frequency hearing, will help estimate the ranges at which certain UCTs can be detected by these species. Detection range estimations, in turn, will allow fisheries researchers to better understand how fish survivorship data obtained using UCTs may be skewed by pinniped predation.

The effect of signal duration on the underwater detection thresholds of a harbor porpoise (Phocoena phocoena) for single frequency-modulated tonal signals between 0.25 and 160 kHz

The underwater hearing sensitivity of a young male harbor porpoise for tonal signals of various signal durations was quantified by using a behavioral psychophysical technique. The animal was trained to respond only when it detected an acoustic signal. Fifty percent detection thresholds were obtained for tonal signals (15 frequencies between 0.25-160 kHz, durations 0.5-5000 ms depending on the frequency; 134 frequency-duration combinations in total). Detection thresholds were quantified by varying signal amplitude by the 1-up 1-down staircase method. The hearing thresholds increased when the signal duration fell below the time constant of integration. The time constants, derived from an exponential model of integration [Plomp and Bouman, J. Acoust. Soc. Am. 31, 749-758 (1959)], varied from 629 ms at 2 kHz to 39 ms at 64 kHz. The integration times of the porpoises were similar to those of other mammals including humans, even though the porpoise is a marine mammal and a hearing specialist. The results enable more accurate estimations of the distances at which porpoises can detect short-duration environmental tonal signals. The audiogram thresholds presented by Kastelein et al. [J. Acoust. Soc. Am. 112, 334-344 (2002)], after correction for the frequency bandwidth of the FM signals, are similar to the results of the present study for signals of 1500 ms duration. Harbor porpoise hearing is more sensitive between 2 and 10 kHz, and less sensitive above 10 kHz, than formerly believed.

VARIATION IN ACOUSTIC SIGNALLING TRAITS EXHIBITS FOOTPRINTS OF SEXUAL SELECTION

Phenotypic variation is ubiquitous in nature and a precondition for adaptive evolution. However, theory predicts that the extent of phenotypic variation should decrease with increasing strength of selection on a trait. Comparative analyses of trait variability have repeatedly used this expectation to infer the type or strength of selection. Yet, the suggested influence of selection on trait variability has rarely been tested empirically. In the present study, I compare estimates of sexual selection strength and trait variability from published data. I constricted the analysis to acoustic courtship traits in amphibians and insects with known variability and corresponding results of female binary choice experiments on these traits. Trait variability and strength of sexual selection were significantly correlated, and both were correlated with signal duration. Because traits under stronger selection had lower variation even after the effect of signal duration was eliminated, I conclude that traces of the strength of selection can be observed with respect to variation of acoustic signaling traits in insects and amphibians. The analysis also shows that traits under stabilizing selection have significantly lower phenotypic variability than traits under directional selection.

The effect of signal duration on the underwater hearing thresholds of two harbor seals (Phoca vitulina) for single tonal signals between 0.2 and 40 kHz

The underwater hearing sensitivities of two 2-year-old female harbor seals were quantified in a pool built for acoustic research by using a behavioral psycho-acoustic technique. The animals were trained only to respond when they detected an acoustic signal ("go/no-go" response). Detection thresholds were obtained for pure tone signals (frequencies: 0.2-40 kHz; durations: 0.5-5000 ms, depending on the frequency; 59 frequency-duration combinations). Detection thresholds were quantified by varying the signal amplitude by the 1-up, 1-down staircase method, and were defined as the stimulus levels, resulting in a 50% detection rate. The hearing thresholds of the two seals were similar for all frequencies except for 40 kHz, for which the thresholds differed by, on average, 3.7 dB. There was an inverse relationship between the time constant (tau), derived from an exponential model of temporal integration, and the frequency [log(tau)=2.86-0.94 log(f);tau in ms and f in kHz]. Similarly, the thresholds increased when the pulse was shorter than approximately 780 cycles (independent of the frequency). For pulses shorter than the integration time, the thresholds increased by 9-16 dB per decade reduction in the duration or number of cycles in the pulse. The results of this study suggest that most published hearing thresholds <or=1 kHz for harbor seals are probably not absolute, as they were derived from signals with durations shorter than the time constants for those frequencies.

Localization of noise bursts in the median plane: Effects of duration and level

Previous research has revealed that human sound localization in the median sagittal and in the front/back dimension may depend on the listener’s ability to reliably extract information regarding characteristic peaks and notches of the source spectrum upon reaching the ears. While long-duration sounds are well localized using spectral cues, the processing of spectral cues for sounds of short duration appears to be degraded. The mechanisms involved in ‘‘image formation’’ in the localization process, and the relative importance of sound duration and level are not well understood. In the present study, the effect of signal duration and level on localization performance, with an emphasis on the median-sagittal plane were systematically explored. Three types of stimuli, clicks, noises, and trains of frozen and random noise bursts encompassing the same time period as the noises, with different duration and at various levels, were randomly presented to the loudspeakers. Subjects were blindfolded, but were instructed to move their heads as necessary, to maximize their confidence regarding source position judgment. Results will be discussed in relation to possible mechanisms employed by the auditory system to utilize spectrally shaped cues in the localization process. [Work supported by NIDCD Grant Nos. R29 DC03083 and P01 DC 00116.]

Effects of signal duration and temporal position on physiological measures of the growth of masking

Psychoacoustic studies have shown that the growth of masking for pure-tone signals in broadband noise is linear when the signal onset is delayed relative to the masker but is nonlinear when signal and masker onsets are synchronous [e.g., Strickland (2001)]. Curiously, for very short duration signals (10 ms), nonlinear effects have been observed for some delayed-onset conditions (Oxenham et al., 1997). While nonlinear effects may result from peripheral compression, it is unclear if other physiological mechanisms might be responsible for the influence of signal temporal parameters on the growth of masking. To address this issue, the growth of masking was studied using evoked-potential responses recorded from the chinchilla inferior colliculus (IC). Quiet and masked thresholds were obtained for 1.0-, 4.0-, and 6.5-kHz signals as a function of signal duration (2–100 ms) and temporal position (delayed or nondelayed) within a 400-ms broadband noise masker. Masker spectrum levels ranged from −10 to 20 dB. Consistent with Stricklands data, physiological growth of masking was linear under delayed conditions and nonlinear under nondelayed conditions. Specifically, for nondelayed conditions, the degree of nonlinearity was greatest for high-frequency signals but did not vary with signal duration. [Work supported by NSF IBN-9996379.]

Effects of signal duration on the recognition of masked communication signals by the grasshopper Chorthippus biguttulus.

The detection and recognition of acoustic communication signals masked by noise was investigated in a grasshopper (Chorthippus biguttulus) whose auditory system exhibits only poor spectral resolution and therefore has to operate in the time domain. The signals of this species consist of numerous identical subunits that enable the receiver, in principle, to make repetitive measurements. We aimed at determining the maximum integration time in this species by using stimuli of different durations under increasing noise levels. As a criterion for recognition the typical phonotactic turning response of the males was evaluated, which is reliably triggered by a female song, and thus is a sensitive indicator for recognition of conspecific signals. When confronted with a long signal (1000 ms) males tolerated a 2.4 dB higher noise level as compared to a short signal (250 ms). Noise tolerance improved with increasing signal duration from 250 ms to 450 ms. Beyond this signal duration, however, no further improvement was observed, indicating an upper limit for temporal integration that corresponds to only five song subunits. The gain in noise tolerance had a slope of 2.7 dB per doubling duration, which corresponds to the expectation derived from an energy detector model (3 dB per doubling duration) rather than to the value expected from signal detection theory (1.5 dB per doubling duration).

Effects of signal duration on analytical frequency meter accuracy

The main factor governing the error in measuring the frequency of a monochromatic signal by means of an analytical frequency meter is the finite signal duration; expressions are derived for the systematic error.

The effects of signal duration on NoSo and NoS pi thresholds at 500 Hz and 4 kHz.

A two-interval, two-alternative temporal forced-choice procedure was used to measure NoSo and NoS pi masked thresholds with 500-Hz and 4-kHz tonal signals. The duration of the signal was either 10, 20, 40, or 320 ms. The maskers were 200-Hz-wide bands of Gaussian noise centered at the frequency of the signal and presented continuously. Decreasing the duration of the 500-Hz tonal signal resulted in a modest increase (1.5 dB or so) in the masking-level difference (MLD) measured between NoSo and NoS pi conditions. In contrast, decreasing the duration of the 4-kHz tonal signal resulted in a substantial decrease (4.5 dB or so) in the MLD. Comparisons of the data with thresholds predicted from analyses based on "windows of temporal integration" provided quantitatively acceptable accounts of the data. The data obtained in the NoS pi condition at 4 kHz, which are novel and were of primary interest, were well-accounted for in a statistical sense. However, there were small, but systematic, discrepancies between the predictions and the data. Those discrepancies, although small in magnitude, suggest that binaural temporal integration at high frequencies, where the envelopes of the stimuli convey the information, may be inherently different from both monaural temporal integration and binaural temporal integration at low frequencies.

Effect of signal duration on categorical loudness scaling in normal and in hearing-impaired listeners.

The present study sought to determine whether the duration of white-noise bursts affects their loudness category rating in the same way for hearing-impaired as for normally-hearing subjects. Twelve normally-hearing and 12 hearing-impaired subjects took part. Categorical loudness growth functions were obtained for 16.25 ms, 32.5 ms, 75 ms, 150 ms and 300 ms white noise bursts. Temporal integration of loudness was defined as the intensity difference needed for stimuli of different durations to result in identical category ratings. In normally-hearing subjects, temporal integration of loudness occurred mainly with the short-duration (16.25 ms and 32.5 ms) stimuli, whereas it was found with almost every stimulus duration in hearing-impaired subjects. In other words, temporal integration of loudness between 16.25 ms and 300 ms stimulus duration was greater in hearing-impaired listeners and there was a difference between normal and hearing-impaired subjects regarding change in loudness perception with stimulus duration. Consequently, the use of fixed-duration stimuli hinders loudness normalization.

Effects of signal duration on vibrotactile intensity discrimination

Vibrotactile intensity DLs were measured by the continuous pedestal and gated pedestal methods. In both cases, the relative DL decreased as a function of the sensation level of the stimulus and the results, in most cases, could be described as a near miss to Weber’s law. DLs measured by the continuous pedestal method were found to decrease substantially as a function of increases in stimulus duration over a range of 12 to 1000 ms and the results could be described by a temporal summation function derived from Zwislocki’s theory of temporal summation [J. J. Zwislocki, J. Acoust. Soc. Am. 32, 1046–1060 (1960)]. In contrast, DLs measured by the gated pedestal method were only slightly affected by changes in stimulus duration. These findings are consistent with the hypothesis that, at suprathreshold levels of stimulation, the effects of temporal summation tend to be cancelled by the effects of adaptation.

Contingency theory and the effect of the duration of signals for noncontingent reinforcement

Rats learned to approach a light that signaled food when food occurred only after the signal, but did not approach the light when food was equally probable regardless of the signal. When food unpaired with the light was preceded by a 5-sec noise, little learning was evident as well, but when the unpaired food was preceded by a 15-sec noise, approach to the light conditioned stimulus readily occurred. These effects of signal duration are similar to those previously obtained with the effect of free food on free-operant behavior and challenge existing theories of contingency effects in Pavlovian conditioning.

Detection of unexpected tones in noise: The effect of signal duration

The effect of signal duration on the detectability of unexpected tones masked by a continuous wideband noise was assessed using a probe-signal method. This method encourages the listener to attend to a target frequency by presenting the signal most often at that frequency, and only occasionally at other unexpected probe frequencies. The attention contour (percent correct as a function of probe frequency) was considerably broader with 5-ms than with 295-ms signals. However, auditory filter shapes measured using the notched-noise technique were very similar for those two signal durations, indicating that the results obtained in the attention experiments do not simply reflect peripheral frequency selectivity. Further supporting this interpretation, probe tones having the same frequency but a different duration from the target were poorly detected. It is proposed that the subject listens through a time-frequency window whose location and shape in the time-frequency plane is determined by the duration and frequency of the target. [Work supported by the NIDCD and NIH.]

The effect of signal duration on frequency discrimination at low signal-to-noise ratios in different conditions of interaural phase

Frequency discrimination in noise at low signal-to-noise ratios where the signals to be discriminated are equally well detected but only faintly audible is better when the signals are in-phase at the ears than when they are out-of-phase. In spite of the difference in discriminability, however, the form of the dependence of the just-noticeable frequency difference on signal duration is the same in both cases.

Influence of signal duration on the masking-level difference.

The effect of signal duration on the 500-Hz masking-level difference was studied in 4 subjects using an adaptive, two-interval, forced-choice method. SoNo, S pi No, and SoN pi thresholds for 13 signal durations between 2 and 100 ms were established in continuous broadband noise (36.2 dB pressure-spectrum level). The signals were synthesized and gated digitally with 1-ms rise-fall times. The SoNo threshold integration functions diverged from the S pi No or the SoN pi threshold integration functions as the signal duration was reduced from 100 ms to 6-10 ms. As the signal duration was reduced below 6-10 ms, however, the threshold integration functions for SoNo, S pi No, and SoN pi converged. The relationships among the threshold integration functions were reflected in the SoNo-S pi No and the SoNo-SoN pi masking-level differences that increased as the signal duration was reduced from 100 to 6-10 ms and then decreased as the signal duration was reduced below 6-10 ms.