Spectral loudness summation for short and long signals as a function of level

Deep learning and machine learning-based voice analysis for the detection of COVID-19: A proposal and comparison of architectures

Alongside the currently used nasal swab testing, the COVID-19 pandemic situation would gain noticeable advantages from low-cost tests that are available at any-time, anywhere, at a large-scale, and with real time answers. A novel approach for COVID-19 assessment is adopted here, discriminating negative subjects versus positive or recovered subjects. The scope is to identify potential discriminating features, highlight mid and short-term effects of COVID on the voice and compare two custom algorithms. A pool of 310 subjects took part in the study; recordings were collected in a low-noise, controlled setting employing three different vocal tasks. Binary classifications followed, using two different custom algorithms. The first was based on the coupling of boosting and bagging, with an AdaBoost classifier using Random Forest learners. A feature selection process was employed for the training, identifying a subset of features acting as clinically relevant biomarkers. The other approach was centered on two custom CNN architectures applied to mel-Spectrograms, with a custom knowledge-based data augmentation. Performances, evaluated on an independent test set, were comparable: Adaboost and CNN differentiated COVID-19 positive from negative with accuracies of 100% and 95% respectively, and recovered from negative individuals with accuracies of 86.1% and 75% respectively. This study highlights the possibility to identify COVID-19 positive subjects, foreseeing a tool for on-site screening, while also considering recovered subjects and the effects of COVID-19 on the voice. The two proposed novel architectures allow for the identification of biomarkers and demonstrate the ongoing relevance of traditional ML versus deep learning in speech analysis.

Relationship between Behavioral and Objective Measures of Sound Intensity in Normal-Hearing Listeners and Hearing-Aid Users: A Pilot Study

Background: For hearing-impaired individuals, hearing aids are clinically fit according to subjective measures of threshold and loudness. The goal of this study was to evaluate objective measures of loudness perception that might benefit hearing aid fitting. Method: Seventeen adult hearing aid users and 17 normal-hearing adults participated in the study. Outcome measures including categorical loudness scaling, cortical auditory evoked potentials (CAEPs), and pupillometry. Stimuli were 1-kHz tone bursts presented at 40, 60, and 80 dBA. Results: Categorical loudness scaling showed that loudness significantly increased with intensity for all participants (p < 0.05). For CAEPs, high intensity was associated with greater P1, N1, and P2 peak amplitude for all listeners (p < 0.05); a significant but small effect of hearing aid amplification was observed. For all participants, pupillometry showed significant effects of high intensity on pupil dilation (p < 0.05); there was no significant effect of hearing aid amplification. A Focused Principal Component analysis revealed significant correlations between subjective loudness and some of the objective measures. Conclusion: The present data suggest that intensity had a significant impact on loudness perception, CAEPs, and pupil response. The correlations suggest that pupillometry and/or CAEPs may be useful in determining comfortable amplification for hearing aids.

Uni- and bilateral spectral loudness summation and binaural loudness summation with loudness matching and categorical loudness scaling

OBJECTIVE

Current hearing aid prescription rules assume that spectral loudness summation decreases with hearing impairment and that binaural loudness summation is independent of hearing loss and signal bandwidth. Previous studies have shown that these assumptions might be incorrect. Spectral loudness summation was measured and compared for loudness scaling and loudness matching.

DESIGN

In this study, the effect of bandwidth on binaural summation was investigated by comparing loudness perception of low-pass filtered, high-pass filtered, and broadband pink noise at 35 Categorical Units for both unilateral and bilateral presentation.

STUDY SAMPLE

Sixteen hearing-impaired listeners.

RESULTS

The results show that loudness differences between the three signals are different for bilateral presentation than for unilateral presentation. In specific, binaural loudness summation is larger for the low-pass filtered pink noise than for the high-pass filtered pink noise. Finally, individual variability in loudness perception near loudness discomfort level was found to be very large.

CONCLUSIONS

Loudness matching is offered as a fast and reliable method to measure individual loudness perception. As discomfort with loud sounds is one of the major problems encountered by hearing aid users, measurement of individual loudness perception could improve hearing aid fitting substantially.

Partial loudness at masker onset indicates temporal effects at supra-threshold levels

This study examines temporal effects both at threshold and at supra-threshold levels. The level needed to detect a short-duration 4.0-kHz signal was measured for signals presented with different onset delays relative to a 300-ms broadband noise masker: 100 ms and 5 ms before the onset of the masker and 5 ms and 100 ms after the onset of the masker. Loudness matches between the signal in quiet and the signal at the same four onset delays were obtained for five presentation levels of the short-duration signal and for three masker levels. The temporal effect was defined as the level difference between the signals near masker onset and the signals well before or well after masker onset, needed to reach threshold and/or achieve equal loudness. Both at threshold and at supra-threshold levels temporal effects were observed consistent with a decrease in gain at the masker frequency during the course of the masker. The temporal effect was not restricted to simultaneous masking, but was also found for backward masking. In both cases the temporal effects were stronger at supra-threshold levels than at threshold. This may be caused by a transient effect at masker onset. The almost simultaneous onset of the signal and the masker makes it difficult for subjects to separate signal from the masker, especially when the signal level is close to masked threshold.

Benefit of binaural listening as revealed by speech intelligibility and listening effort

In contrast to the well-known benefits for speech intelligibility, the advantage afforded by binaural stimulus presentation for reducing listening effort has not been thoroughly examined. This study investigated spatial release of listening effort and its relation to binaural speech intelligibility in listeners with normal hearing. Psychometric functions for speech intelligibility of a frontal target talker masked by a stationary speech-shaped noise were estimated for several different noise azimuths, different degrees of reverberation, and by maintaining only interaural level or time differences. For each of these conditions, listening effort was measured using a categorical scaling procedure. The results revealed that listening effort was significantly reduced when target and masker were spatially separated in anechoic conditions. This effect extended well into the range of signal-to-noise ratios (SNRs) in which speech intelligibility was at ceiling, and disappeared only at the highest SNRs. In reverberant conditions, spatial release from listening effort was observed for high, but not low, direct-to-reverberant ratios. The findings suggest that listening effort assessment can be a useful method for revealing the benefits of spatial separation of sources under realistic listening conditions comprising favorable SNRs and low reverberation, which typically are not apparent by other means.

Influence of suppression on restoration of spectral loudness summation in listeners with hearing loss

Loudness depends on both the intensity and spectrum of a sound. Listeners with normal hearing perceive a broadband sound as being louder than an equal-level narrowband sound because loudness grows nonlinearly with level and is then summed across frequency bands. This difference in loudness as a function of bandwidth is reduced in listeners with sensorineural hearing loss (SNHL). Suppression, the reduction in the cochlear response to one sound by the simultaneous presentation of another sound, is also reduced in listeners with SNHL. Hearing-aid gain that is based on loudness measurements with pure tones may fail to restore normal loudness growth for broadband sounds. This study investigated whether hearing-aid amplification that mimics suppression can improve loudness summation for listeners with SNHL. Estimates of loudness summation were obtained using measurements of categorical loudness scaling (CLS). Stimuli were bandpass-filtered noises centered at 2 kHz with bandwidths in the range of 0.1-6.4 kHz. Gain was selected to restore normal loudness based on CLS measurements with pure tones. Gain that accounts for both compression and suppression resulted in better restoration of loudness summation, compared to compression alone. However, restoration was imperfect, suggesting that additional refinements to the signal processing and gain-prescription algorithms are needed.

Deriving loudness growth functions from categorical loudness scaling data

The goal of this study was to reconcile the differences between measures of loudness obtained with continuous, unbounded scaling procedures, such as magnitude estimation and production, and those obtained using a limited number of discrete categories, such as categorical loudness scaling (CLS). The former procedures yield data with ratio properties, but some listeners find it difficult to generate numbers proportional to loudness and the numbers cannot be compared across listeners to explore individual differences. CLS, where listeners rate loudness on a verbal scale, is an easier task, but the numerical values or categorical units (CUs) assigned to the points on the scale are not proportional to loudness. Sufficient CLS data are now available to assign values in sones, a scale proportional to loudness, to the loudness categories. As a demonstration of this approach, data from Heeren, Hohmann, Appell, and Verhey [J. Acoust. Soc. Am. 133, EL314-EL319 (2013)] were used to develop a CUsone metric, whose values were then substituted for the original CU values in reanalysis of a large set of CLS data obtained by Rasetshwane, Trevino, Gombert, Liebig-Trehearn, Kopun, Jesteadt, Neely, and Gorga [J. Acoust. Soc. Am. 137, 1899-1913 (2015)]. The resulting data are well fitted by power functions and are in general agreement with previously published results obtained with magnitude estimation, magnitude production, and cross modality matching.

Spectral and binaural loudness summation for hearing-impaired listeners

Sensorineural hearing loss typically results in a steepened loudness function and a reduced dynamic range from elevated thresholds to uncomfortably loud levels for narrowband and broadband signals. Restoring narrowband loudness perception for hearing-impaired (HI) listeners can lead to overly loud perception of broadband signals and it is unclear how binaural presentation affects loudness perception in this case. Here, loudness perception quantified by categorical loudness scaling for nine normal-hearing (NH) and ten HI listeners was compared for signals with different bandwidth and different spectral shape in monaural and in binaural conditions. For the HI listeners, frequency- and level-dependent amplification was used to match the narrowband monaural loudness functions of the NH listeners. The average loudness functions for NH and HI listeners showed good agreement for monaural broadband signals. However, HI listeners showed substantially greater loudness for binaural broadband signals than NH listeners: on average a 14.1 dB lower level was required to reach "very loud" (range 30.8 to -3.7 dB). Overall, with narrowband loudness compensation, a given binaural loudness for broadband signals above "medium loud" was reached at systematically lower levels for HI than for NH listeners. Such increased binaural loudness summation was not found for loudness categories below "medium loud" or for narrowband signals. Large individual variations in the increased loudness summation were observed and could not be explained by the audiogram or the narrowband loudness functions.

Noise trauma induced plastic changes in brain regions outside the classical auditory pathway

The effects of intense noise exposure on the classical auditory pathway have been extensively investigated; however, little is known about the effects of noise-induced hearing loss on non-classical auditory areas in the brain such as the lateral amygdala (LA) and striatum (Str). To address this issue, we compared the noise-induced changes in spontaneous and tone-evoked responses from multiunit clusters (MUC) in the LA and Str with those seen in auditory cortex (AC) in rats. High-frequency octave band noise (10-20 kHz) and narrow band noise (16-20 kHz) induced permanent threshold shifts at high-frequencies within and above the noise band but not at low frequencies. While the noise trauma significantly elevated spontaneous discharge rate (SR) in the AC, SRs in the LA and Str were only slightly increased across all frequencies. The high-frequency noise trauma affected tone-evoked firing rates in frequency and time-dependent manner and the changes appeared to be related to the severity of noise trauma. In the LA, tone-evoked firing rates were reduced at the high-frequencies (trauma area) whereas firing rates were enhanced at the low-frequencies or at the edge-frequency dependent on severity of hearing loss at the high frequencies. The firing rate temporal profile changed from a broad plateau to one sharp, delayed peak. In the AC, tone-evoked firing rates were depressed at high frequencies and enhanced at the low frequencies while the firing rate temporal profiles became substantially broader. In contrast, firing rates in the Str were generally decreased and firing rate temporal profiles become more phasic and less prolonged. The altered firing rate and pattern at low frequencies induced by high-frequency hearing loss could have perceptual consequences. The tone-evoked hyperactivity in low-frequency MUC could manifest as hyperacusis whereas the discharge pattern changes could affect temporal resolution and integration.

Categorical scaling of partial loudness in a condition of masking release

Categorical loudness scaling was used to measure suprathreshold release from masking. The signal was a 986-Hz sinusoid that was embedded in a bandpass-filtered masking noise. This noise was either unmodulated or was amplitude modulated with a square-wave modulator. The unmodulated noise had either the same level as the modulated noise or had a level that was reduced by the difference in thresholds for the 986-Hz signal obtained with the modulated and unmodulated noise masker presented at the same level (i.e., the masking release). A comparison with loudness matching data of the same set of subjects showed that the data obtained with loudness scaling capture main aspects of the change in suprathreshold perception of the sinusoid when the masker was modulated. The scaling data for the signal masked by the unmodulated noise with the reduced masker level were similar to that for the signal embedded in the modulated noise. This similarity supports the hypothesis that the mechanism eliciting the masking release is effectively reducing the masker level.

Categorical loudness scaling and equal-loudness contours in listeners with normal hearing and hearing loss

This study describes procedures for constructing equal-loudness contours (ELCs) in units of phons from categorical loudness scaling (CLS) data and characterizes the impact of hearing loss on these estimates of loudness. Additionally, this study developed a metric, level-dependent loudness loss, which uses CLS data to specify the deviation from normal loudness perception at various loudness levels and as function of frequency for an individual listener with hearing loss. CLS measurements were made in 87 participants with hearing loss and 61 participants with normal hearing. An assessment of the reliability of CLS measurements was conducted on a subset of the data. CLS measurements were reliable. There was a systematic increase in the slope of the low-level segment of the CLS functions with increase in the degree of hearing loss. ELCs derived from CLS measurements were similar to standardized ELCs (International Organization for Standardization, ISO 226:2003). The presence of hearing loss decreased the vertical spacing of the ELCs, reflecting loudness recruitment and reduced cochlear compression. Representing CLS data in phons may lead to wider acceptance of CLS measurements. Like the audiogram that specifies hearing loss at threshold, level-dependent loudness loss describes deficit for suprathreshold sounds. Such information may have implications for the fitting of hearing aids.

Loudness of subcritical sounds as a function of bandwidth, center frequency, and level

Level differences at equal loudness between band-pass noise and pure tones with a frequency equal to the center frequency of the noise were measured in normal-hearing listeners using a loudness matching procedure. The center frequencies were 750, 1500, and 3000 Hz and noise bandwidths from 5 to 1620 Hz were used. The level of the reference pure tone was 30, 50, or 70 dB. For all center frequencies and reference levels, the level at equal loudness was close to 0 dB for the narrowest bandwidth, increased with bandwidth for bandwidths smaller than the critical bandwidth, and decreased for bandwidths larger than the critical bandwidth. For bandwidths considerably larger than the critical bandwidth, the level difference was negative. The maximum positive level difference was measured for a bandwidth close to the critical bandwidth. This maximum level difference decreased with increasing reference level. A similar effect was found when the level differences were derived from data of an additional categorical loudness scaling experiment. The results indicate that the decrease of loudness at equal level with increasing subcritical bandwidth is a common property of the auditory system which is not taken into account in current loudness models.

Relation between loudness in categorical units and loudness in phons and sones

Data are presented on the relation between loudness measured in categorical units (CUs) using a standardized loudness scaling method (ISO 16832, 2006) and loudness expressed as the classical standardized measures phon and sone. Based on loudness scaling of narrowband noise signals by 31 normal-hearing subjects, sound pressure levels eliciting the same categorical loudness were derived for various center frequencies. The results were comparable to the standardized equal-loudness level contours. A comparison between the loudness function in CUs at 1000 Hz and the standardized loudness function in sones indicates a cubic relation between the two loudness measures.

Spectro-temporal weighting of loudness

Real-world sounds like speech or traffic noise typically exhibit spectro-temporal variability because the energy in different spectral regions evolves differently as a sound unfolds in time. However, it is currently not well understood how the energy in different spectral and temporal portions contributes to loudness. This study investigated how listeners weight different temporal and spectral components of a sound when judging its overall loudness. Spectral weights were measured for the combination of three loudness-matched narrowband noises with different center frequencies. To measure temporal weights, 1,020-ms stimuli were presented, which randomly changed in level every 100 ms. Temporal weights were measured for each narrowband noise separately, and for a broadband noise containing the combination of the three noise bands. Finally, spectro-temporal weights were measured with stimuli where the level of the three narrowband noises randomly and independently changed every 100 ms. The data consistently showed that (i) the first 300 ms of the sounds had a greater influence on overall loudness perception than later temporal portions (primacy effect), and (ii) the lowest noise band contributed significantly more to overall loudness than the higher bands. The temporal weights did not differ between the three frequency bands. Notably, the spectral weights and temporal weights estimated from the conditions with only spectral or only temporal variability were very similar to the corresponding weights estimated in the spectro-temporal condition. The results indicate that the temporal and the spectral weighting of the loudness of a time-varying sound are independent processes. The spectral weights remain constant across time, and the temporal weights do not change across frequency. The results are discussed in the context of current loudness models.

Restoration of loudness summation and differential loudness growth in hearing-impaired listeners

When normal-hearing (NH) listeners compare the loudness of narrowband and wideband sounds presented at identical sound pressure levels, the wideband sound will most often be perceived as louder than the narrowband sound, a phenomenon referred to as loudness summation. Hearing-impaired (HI) listeners typically show less-than-normal loudness summation, due to reduced cochlear compressive gain and degraded frequency selectivity. In the present study, loudness summation at 1 and 3 kHz was estimated monaurally for five NH and eight HI listeners by matching the loudness of narrowband and wideband noise stimuli. The loudness summation was measured as a function both of noise bandwidth and level. The HI listeners were tested unaided and aided using three different compression systems to investigate the possibility of restoring loudness summation in these listeners. A compression system employing level-dependent compression channels yielded the most promising outcome. The present results inform the development of future loudness models and advanced compensation strategies for the hearing impaired.

Spectral loudness summation of nonsimultaneous tone pulses

The level of broadband signals is usually lower than that of equally loud narrow-band signals. This effect, referred to as spectral loudness summation, is commonly measured for broadband signals where all frequency components are presented simultaneously. The present study investigated to what extent spectral loudness summation also occurs for nonsimultaneously presented frequency components. Spectral loudness summation was measured in normal-hearing listeners with an adaptive forced-choice procedure for sequences of short tone pulses with varying frequencies, randomly chosen from a set of five frequencies. In addition, spectral loudness summation was measured for the simultaneous presentation of all five frequencies. The comparison stimulus consisted of tone pulses with the same frequency for all tone pulses of the sequence and the same repetition rate and overall duration as the test signal. The pulse duration was 10, 20, 50, or 100 ms and the inter-pulse interval ranged from 0 to 390 ms. In general, a considerable nonsimultaneous spectral loudness summation was found for short pulse durations and inter-pulse intervals, but a residual effect was also observed for the largest inter-pulse interval. The data are discussed in the light of repetition-rate dependent spectral loudness summation and effects of persistence of specific loudness after tone-pulse offset.

Temporal integration of loudness measured using categorical loudness scaling and matching procedures

Temporal integration of loudness of 1 kHz tones with 5 and 200 ms durations was assessed in four subjects using two loudness measurement procedures: categorical loudness scaling (CLS) and loudness matching. CLS provides a reliable and efficient procedure for collecting data on the temporal integration of loudness and previously reported nonmonotonic behavior observed at mid-sound pressure level levels is replicated with this procedure. Stimuli that are assigned to the same category are effectively matched in loudness, allowing the measurement of temporal integration with CLS without curve-fitting, interpolation, or assumptions concerning the form of the loudness growth function.

Assessment of auditory nonlinearity for listeners with different hearing losses using temporal masking and categorical loudness scaling

A dysfunction or loss of outer hair cells (OHC) and inner hair cells (IHC), assumed to be present in sensorineural hearing-impaired listeners, affects the processing of sound both at and above the listeners' hearing threshold. A loss of OHC may be responsible for a reduction of cochlear gain, apparent in the input/output function of the basilar membrane and steeper-than-normal growth of loudness with level (recruitment). IHC loss is typically assumed to cause a level-independent loss of sensitivity. In the current study, parameters reflecting individual auditory processing were estimated using two psychoacoustic measurement techniques. Hearing loss presumably attributable to IHC damage and low-level (cochlear) gain were estimated using temporal masking curves (TMC). Hearing loss attributable to OHC (HL(OHC)) was estimated using adaptive categorical loudness scaling (ACALOS) and by fitting a loudness model to measured loudness functions. In a group of listeners with thresholds ranging from normal to mild-to-moderately impaired, the loss in low-level gain derived from TMC was found to be equivalent with HL(OHC) estimates inferred from ACALOS. Furthermore, HL(OHC) estimates obtained using both measurement techniques were highly consistent. Overall, the two methods provide consistent measures of auditory nonlinearity in individual listeners, with ACALOS offering better time efficiency.

Temporal weighting in loudness of broadband and narrowband signals

Temporal weights used by listeners when judging the overall loudness of a stimulus were measured for a 1-s-long noise centered around 2 kHz, whose level was randomly perturbed every 100 ms. The bandwidth was either 6400 Hz (broadband condition) or 400 Hz (narrowband condition). The first 100 ms contributed significantly more than later segments to overall loudness perception in the broadband condition. The effect was significantly reduced in the narrowband condition which is in line with the hypothesis that a greater spectral loudness summation at stimulus onset might be the mechanism behind the onset accentuation.

Spectral loudness summation for sequences of short noise bursts

Recent loudness data of single noise bursts indicate that spectral loudness summation depends on signal duration. To gain insight into the mechanisms underlying this duration effect, loudness was measured as a function of signal bandwidth centered around 2 kHz for sequences of 10-ms noise bursts at various repetition rates and, for comparison, for single noise bursts of either 10- or 1000-ms duration. The test-signal bandwidth was varied from 200 to 6400 Hz. For the repeated noise bursts, the reference signal had a bandwidth of 400 Hz. For the single noise bursts, data were obtained for two reference bandwidths: 400 and 3200 Hz. In agreement with previous results, the magnitude of spectral loudness summation was larger for the 10-ms than for the 1000-ms noise bursts. The reference bandwidth had no significant effect on the results for the single noise bursts. Up to repetition rates of 50 Hz, the magnitude of spectral loudness summation for the sequences of noise bursts was the same as for the single short noise burst. The data indicate that the mechanism underlying the duration effect in spectral loudness is considerably faster than the time constant of about 100 ms commonly associated with the temporal integration of loudness.