Recent Abstracts
From the 36th midwinter meeting of the Association for Research in Otolaryngolgoy (2013):
Glimpsing a Target Harmonic Complex in a Temporally Interrupted Masker
Yi Shen
University of California, Irvine
Background
When two sounds (a target and a masker) are presented simultaneously, listeners are typically more successful in identify features in the target sound if the masker sound contains temporal gaps. This benefit could be a consequence of the improved target-to-masker ratio (TMR) during the temporal gaps, or it might be interpreted that the temporal gaps in the masker provide a cue that promotes the perceptual segregation of the two sounds.
Methods
In the current study, sensitivity to changes in spectral profile was measured for a harmonic complex target sound. The target was always presented with a simultaneous harmonic masker that was temporally interrupted. Performance thresholds were collected as functions of differences in fundamental frequency (from 0 to 10 semitones) and target-to-masker ratio (from -20 to 10 dB). Besides thresholds, relative decision weights were estimated independently for the masker, the portions of the target that temporally overlapped the masker (target-overlap), and the portions of the target that were in the masker gaps (target-gap).
Results
Profile analysis thresholds improved with increasing fundamental-frequency difference, TMR, and gap duration. At low TMR’s, the target-gap stimulus dominated the responses whereas the target-overlap contributed little to the responses, indicating that listeners conducted the task by glimpsing the target in the masker gaps. At high target-to-masker ratios, the target-overlap and target-gap contributed equally to the responses.
Conclusions
When it is advantageous to do so (e.g., at low TMR’s), listeners perform profile analysis of the target using glimpses of the target during the temporal gaps of the concurrent masker. When the gap duration is short and the fundamental-frequency difference between the target and masker is small, the masker cannot be totally ignored. The efficiency of glimpsing the target improved as the fundamental-frequency difference increases.
From the 36th midwinter meeting of the Association for Research in Otolaryngolgoy (2013):
Efficient Parameter-Based Estimates of Auditory Filters
Yi Shen and Virginia M. Richards
University of California, Irvine
Background
Performance-based methods currently used to estimate auditory filter parameters are time-consuming, making routine estimates difficult. An adaptive “parameter-based” Bayesian adaptive procedure for the rapid estimation of auditory filter parameters was proposed and tested.
Methods
The power spectrum model of masking was assumed with an efficiency parameter K and a two-parameter roex (r, p) filter, yielding a 3-D space of model parameters. A logistic psychometric function linked the model to behavioral responses. The task was the detection of a tone added to a notched noise, yielding a 2-D stimulus space composed of potential normalized notch bandwidths and signal strengths in dB SPL. Following Kontsevich and Tyler [1999; Vis. Res. 39, 2729-2737], a one-step-ahead search algorithm with an entropy-based criterion was adopted in which the next stimulus to be tested was the one that decreased the total expected entropy in the parameter estimation.
Results
The procedure was successful for most young normal-hearing listeners in that parameter estimates obtained from the proposed procedure and 100-150 trials were as reliable as those obtained using 1000 trials with a performance-based procedure. For a few listeners, the procedure failed to converge to reasonable parameter estimates, and for them, the performance-based measures were also variable. By introducing a “lapse” term into the psychometric function, the parameter-based procedure converged rapidly (100-150 trials) and reliably even for these few listeners.
Conclusions
The current data provided a proof-of-concept that estimates of auditory filters for young, normal hearing listeners could be achieved in as few as 150 trials. This parameter-based procedure can be extended to include other populations and other well-modeled psychophysical function.
From the 163rd annual meeting of the Acoustical Society of America (2012):
The Use of Relative Weights to Assess Perceptual Segregation in a Concurrent Profile Analysis Task
Yi Shen
University of California, Irvine
A series of experiments were conducted to address the need of a psychophysical tool to measure the perceptual segregation of concurrent sources. The experiments measured listeners’ sensitivity to the spectral profile of a target sound embedded in a concurrent masker. Both the target and masker were harmonic complexes, which were presented at different fundamental frequencies in order to investigate the effects of this acoustic cue on segregation. The task was designed so that it either strongly encouraged the segregation of the two complexes (task-driven design) or it did not necessarily require segregation (listener-driven design). In both cases, the degree of segregation was assessed by deriving the relative decision weights on the target and masker. Larger differences between the target and masker weights were found as the fundamental frequency difference between the two complexes increased (0.5 – 15 semitones), suggesting that listeners were more successful in selectively attending to the target alone at larger fundamental frequency separations. Although quite different thresholds were obtained for the task-driven and listener-driven designs, the estimates of the decision weights were consistent across the two task designs, indicating that listeners’ motivations did not influence the usefulness of the periodicity cue in segregating concurrent sounds.[Supported by NIDCD]
From the 35th annual meeting of the Association for Research in Otolaryngology (2012):
Decision Processes in Cued Informational Masking Experiments
Virginia M. Richards and Eva Maria Carreira
University of California, Irvine
When a target sinusoid is masked by a random multitone informational masker, thresholds are lower when a preview of the masker precedes the detection interval than when a signal-plus-masker preview precedes the detection interval. In this experiment the decision processes associated with these two types of “cues” were examined. Using a yes/no procedure, the task was to detect a 1000-Hz tone presented concurrently with an informational masker. For the cue and detection interval the number of tones that comprised the masker was independently drawn at random to be 6, 7, or 8. Then, for the cue interval, the frequencies of the masker components were randomly chosen. If the number of components was the same for the masker in the cue and detection interval, the same frequencies were used; if not, the interval with the larger number of masker components used the same components presented in the other interval, plus any additional components with newly chosen random frequencies. The analyses of the data from a two-down, one-up tracking procedure was analyzed to determine the degree to which extra masker tones altered decision strategies. The results indicated that (a) when a masker cue was tested, extra masker components in the cue interval had little effect on responses, but for the detection interval extra components in the region of the signal frequency tended to be associated with “signal” responses; and (b) when a signal-plus-masker cue was tested, extra masker components in the cue interval were associated with “no-signal” responses, and extra components in the detection interval had no measurable impact on responses. These results are not consistent with a model in which decision variables from the cue and detection intervals are compared, and also indicate that listeners adopt different strategies depending on the type of cue tested.[Supported by NIDCD]
From the 35th annual meeting of the Association for Research in Otolaryngology (2012):
Behavioral Assessment of Perceptual Segregation During a Profile Analysis Task Under Simultaneous Masking
Yi Shen and Virginia M. Richards
University of California, Irvine
A narrow-band signal is subjective to less masking from a simultaneous notched masker if it is preceded by a precursor that occupies the same spectral region as the masker. The present study investigates the effect of the precursor on the internal representations of the signal and masker by measuring the masking pattern for short-duration probes as a function of the spectrotemporal placement. In the enhancement measurements, the signal was a 0.6-octave noise centered around 1 kHz, and the masker and precursor were broadband noises with spectral notches at the signal frequency. The amount of enhancement was measured as a function of the temporal separation between the precursor and masker. For corresponding conditions, the detection thresholds of a 6-ms tonal probe were measured at several temporal locations and at frequency regions occupied by either the signal or masker. These measured masking patterns were compared against the amount of enhancement, which showed that the presence of the precursor altered the internal representations of both the signal and masker. These masking patterns provide an opportunity to generate hypotheses on the plausible mechanisms underlying the auditory enhancement phenomenon.[Supported by NIDCD]
From the 161st meeting of the Acoustical Society of America (2011):
Masking Patterns of Stimuli Exhibiting Enhanced Detection for Spectrally Notched Precursors
Yi Shen and Virginia M. Richards
University of California, Irvine
A narrow-band signal is subjective to less masking from a simultaneous notched masker if it is preceded by a precursor that occupies the same spectral region as the masker. The present study investigates the effect of the precursor on the internal representations of the signal and masker by measuring the masking pattern for short-duration probes as a function of the spectrotemporal placement. In the enhancement measurements, the signal was a 0.6-octave noise centered around 1 kHz, and the masker and precursor were broadband noises with spectral notches at the signal frequency. The amount of enhancement was measured as a function of the temporal separation between the precursor and masker. For corresponding conditions, the detection thresholds of a 6-ms tonal probe were measured at several temporal locations and at frequency regions occupied by either the signal or masker. These measured masking patterns were compared against the amount of enhancement, which showed that the presence of the precursor altered the internal representations of both the signal and masker. These masking patterns provide an opportunity to generate hypotheses on the plausible mechanisms underlying the auditory enhancement phenomenon.[Supported by NIDCD]
From the 35th meeting of the Association for Research in Otolaryngology (2012):
Loudness Matching for Enhanced Sinusoids
Yi Shen
University of California, Irvine
This study concerns the segregation of two concurrent harmonic sounds. Unlike many previous studies, segregation is assessed by modeling listening strategies in addition to measurements of task performance. Sensitivity to changes in spectral envelopes (or spectral profiles) was measured for a target harmonic complex in a competing masker complex. Each of the two complexes contained a single formant in the spectral profile at 1 kHz, varying in its peakiness (or profile strength). In a 2-alternative, forced choice design, the target had low profile strength in one interval and high profile strength in the other interval; while the masker had low profile strength in the both intervals. Small perturbations were applied to the profile strength of the two complexes independently in each interval, from which relative decision weights for each of the two sounds were estimated. Listeners were instructed to select the interval with larger target profile strength. Performance thresholds as well as relative weights were measured as functions of fundamental frequency difference (0 – 6 semitones) and relative intensity (-10, 0, and 10 dB SNR) between the target and masker. Results from 6 normal-hearing listeners indicated that both of these two acoustic features assisted segregation. The effect of fundamental frequency difference was not measurable unless the target-to-masker ratio was low. [Supported by NIDCD]
From the 161st meeting of the Acoustical Society of America (2011):
Concurrent Profile Analysis: The Effect of Segregation Cues on Spectral-Shape Comparisons for Simultaneous Stimuli
Yi Shen and Virginia M. Richards
University of California, Irvine
The ability to accurately perceive spectral information from one sound source in the presence of others is crucial for speech understanding in complex and challenging environments. In the present study, non-speech stimuli, a pair of concurrent complex tones, were used to obtain psychophysical measurements of this ability in a controlled setting. Listeners detected whether the spectral profiles of the two concurrent complexes were the same, or exchanged across two observation intervals. Concurrent profile analysis thresholds were measured as the minimum amount of spectral alterations for the detection of the change. Thresholds were measured from four normal hearing listeners and as functions of three types of segregation cues: difference in fundamental frequency, difference in interaural time differences, and onset asynchrony. The roles of these cues in concurrent spectral-shape analysis are discussed in term of the underlying psychometric functions.[Supported by NIDCD]
From the 34th midwinter meeting of the Association for Research in Otolaryngology (2011):
Contextual Effects in the Identification of Nonspeech Auditory Patterns in Random Maskers
Gerald Kidd Jr., Virginia M. Richards, Timothy Streeter, Christine Mason, and Rong Huang
Boston University and University of California, Irvine
This study investigated the benefit of cuing for the identification of nonspeech auditory patterns in masked conditions. Six target patterns were formed by sequences of four brief sinusoidal tones drawn from narrow frequency bands. Maskers were also four-tone sequences in narrow bands with the individual tones chosen randomly on every trial. The frequency range from 200 to 6540 Hz was divided into 16 equally spaced bands. On any trial, the target occupied one of four possible bands and the maskers occupied eight others. The masker bands were randomized on every trial excluding the target band and the two bands adjacent to the target. In a block of trials, the target pattern was presented in either the same frequency band (fixed) throughout, or was randomized among the four possible bands (random). No significant difference in performance was found for fixed versus random target frequency (with no cues). In three other conditions with random target frequency, various cues were presented just prior to the target-plus-masker interval. The presentation of an exact copy of the masker to the same ear provided the greatest benefit. The other two cues were an ipsilateral notched noise that spanned the range of potential masker frequencies excluding the target region, and an exact copy of the masker presented contralaterally. Both of these resulted in smaller, but significant benefits. One likely mechanism underlying these findings is spectrotemporal “enhancement” which results from differential prior stimulation of target and masker frequency regions. Another factor is the correspondence of the cue to the subsequent masker which exerts a beneficial effect for either ipsilateral or contralateral presentation and may help to reduce masker uncertainty.[Supported by NIDCD and AFOSR]
From the 32nd midwinter meeting of the Association for Research in Otolaryngology (2009):
Tone detection in the presence of continuous vs. pulsed
maskers
Virginia M. Richards, Daniel E. Shub, and Rong Huang
Department of Psychology, University of Pennsylvania
The detectability of a tone added to a noise masker is somewhat improved when the noise masker is
turned on before the signal begins and turned off after the signal ends [e.g., Weir et al., J. Acoust.
Soc. Am., 61, 1298-1300 (1997)]. The current study asked whether this finding held when an informational
masker was tested. The expectation was that for an informational masker there would be no advantage of a
continuous masker because the variation in energy across time and frequency would deter the formation of a
useful reference/template against which the signal frequency could be compared. The two informational maskers
tested were sequences of either three 60-ms or 30-ms bursts comprised of, on average, 20 tone pips drawn at
random on a logarithmic scale ranging from 200 to 5000 Hz. The signal to be detected was a sequence of either
three 60- or six 30-ms equal-frequency tone pips, depending on whether the masker was composed of 60- or 30-ms
bursts,respectively. When present, the signal was temporally aligned with the masker bursts. Pink-noise maskers
were also tested. The masker was either presented continuously, or was pulsed on and off with the sequence of
signal pips. Although there were substantial individual differences, subjects' thresholds were at least as good
when the masker was coninuous as when the masker was pulsed, and typically superior when the masker was
presented continuously, for all masker types. [Supported by NIH DC002012]
From the 32nd midwinter meeting of the Association for Research in Otolaryngology (2009):
Effects of spectral variability on monaural azimuthal localization
Daniel E. Shub and Virginia M. Richards
Department of Psychology, University of Pennsylvania
To localize sound, monaural listeners must rely on the information that the head shadow introduces
into the overall level and spectral shape. The overall level and spectral shcape at the ear, however,
depend on both the source location and the overall level and spectral shape of the source. To investigate
the effects of spectral variability on monaural localization, we measured the ability of normal-hearing
individuals listening monaurally to discriminate between two virtual source locations (± 30 ° from
the midline). The stimulus had a random overall level and a random spectral shape (250 ms multi-tone stimulus
with 3 components per octave between 500 and 8000 Hz). The covariance matrix of the level of each component can
be written as σE2l+ σL2U, where σE and σL are the standard deviations of the component
and overall level pertubations, respectively, and I and U are the identity and unit matrices, respectively.
The value of σL was fixed at 8 dB and σE was parametrically varied between 0 and 8 dB. The measured
d' systematically decreased as σE increased. Subjects were still able to reliably discriminate (values of
d' near unity)between the two source locations with σE equal to 4 dB (and in some cases 8 dB). Although
there were only small individual differences in sensitivity, the subjects' relative weighting patterns, however,
were nearly independent of σE, indicating that each subject had a stable strategy. The overall efficiency,
which accounts for both non-optimal weighting and internal noise, of the subjects was low suggesting that it is
difficult to use the complicated changes in the spectral shape to determine the location of a sound source.
In spite of the low efficiency, discrimination between two locations was still possible with spectrally variable
stimuli. [Supported by NIH DC002012 and DC009384] (
From the 151st meeting of the Acoutical Society of America (2006):
Response-related patterns
in discrimination of FM narrowband noise
Jinyu Qian and Virginia M. Richards
Department of Psychology, University of Pennsylvania
Noise stimuli were used to investigate weighting patterns for detecting
dynamic frequency changes. For one stimulus set, a tonal glide was
modulated by a narrow band of noise (gliding narrowband noise, GNBN,
condition). For the second set, the glide was preceded and followed
by steady-state tones before modulation (SS condition). The task
was to indicate whether the stimulus had a frequency glide or not.
Weighting patterns (as a function of time) were derived for each
subject based on a linear classification model. First, the deviation
of the instantaneous carrier for each stimulus was calculated and
then weighted by the squared envelope of the stimulus. This was
followed by the linear classifications of the squared-envelope-weighted-deviation-of-instantaneous-carrier
(SEWDOIC) according to the subject’s responses [A. Ahumada,
J. Vis., 2, 121-131 (2002)]. The weighting patterns were consistent
across subjects although individual differences in pattern details
were observed. The comparison between the patterns from the GNBN
and SS conditions indicates the use of dynamic information rather
than only comparing the beginning and ending frequencies, consistent
with Lyzenga et al. [J. Acoust. Soc. Am., 116, 491-501 (2004)].
Similar weighting patterns were also estimated when the center frequency
of the stimuli was random.
From the 29th midwinter meeting of the Association for Research
in Otolaryngology (2006):
Simultaneous Two-Channel
Signal Detection, revisited
Gerald Kidd Jr.1, Virginia M. Richards2, Christine R. Mason1, Frederick
J. Gallun1, and Rong Huang2
1Department of Speech, Language and Hearing Sciences and Hearing
Research Center, Boston University, Boston, MA
2Department of Psychology, University of Pennsylvania
This study examined the ability of listeners to monitor events occurring
simultaneously in two widely spaced frequency regions. The task
was to detect the presence of a low frequency (558 Hz) tone and
a high frequency (1791 Hz) tone presented in a 2- interval 4-alternative
forced-choice procedure. On every trial, each signal was presented
in either the first or second interval independently and the listener
was required to indicate the interval of presentation for each signal.
The level of each signal was varied separately according to an adaptive
tracking procedure to obtain threshold estimates. As a control,
thresholds for both signals were also obtained for single-signal
presentation. Comparison of thresholds in single- and dual-signal
presentation conditions provided an estimate of the costs of monitoring
and responding to events in two frequency channels. Signal thresholds
were obtained in quiet, in double-notch-filtered Gaussian noise,
and in random-frequency multitone maskers. Much larger costs (difference
between dual- and single-tone tasks) were found for the masked conditions,
especially for the multitone masker. These results suggest that
the costs of dividing attention along a particular stimulus dimension
depends on both the resources required to solve the task in each
signal channel and also the resources required to ignore competing
stimuli in nonsignal channels. [Work supported by AFOSR and NIH/NIDCD]
From the 149th meeting of the Acoutical Society of America (2005):
Effect of signal frequency
uncertainty for random multi-burst maskers
Rong Huang and Virginia M. Richards
Department of Psychology, University of Pennsylvania
The detectability of a sequence of equal-frequency tone pips masked
by random multi-burst complexes may depend on the perceptual segregation
of the signal stream from the random masker. If so, detection thresholds
may be independent of whether the signal frequency is known versus
uncertain. In this experiment observers detected a signal stream
of 8 sequential equal-frequency 30 ms tone pips embedded in a random
masker composed of 8 sequential bursts. A yes/no procedure was used,
and the independent variable p (probability a tone was played at
a particular time-by-frequency location) governed the number of
masker tones in each burst. The dependent variable was d. Threshold
values of p were obtained for signal streams at 5 different frequencies.
Sensitivity was superior for the mid-frequency signal, and decreased
as the signal frequencies approached the edge of the frequencies
the masker tones occupied (200--5000 Hz). When the frequency of
the signal stream was randomly varied from trial to trial, sensitivity
was poorer than for any of the fixed-frequency signals. Thus, the
detectability of a sequence of tone pips is reduced when the signal
frequency is uncertain compared to certain. Additionally, sensitivity
increased when the signal stream was delayed relative to the masker
bursts.
From the 149th meeting of the Acoutical Society of America (2005):
Comparing linear regression
models applied to psychophysical data
Zhongzhou Tang1, Andrew Shih2 and Virginia M. Richards1
Dept. of Psychology1, Dept. of Bioengineering2, University of Pennsylvania
Relative weights for a profile analysis task were obtained using
four regression/classification models; correlation coefficients,
linear regression, logistic regression and probit regression. The
aim of the study was to examine the impact of the choice of model
on the accuracy and the efficiency with which the relative weights
were determined. A yes/no task was used with observers indicating
whether or not there was an increment in level to the central component
of an 11-component standard. On each presentation the amplitudes
of the individual components of the complex were randomly perturbed
using draws from a normal distribution. When a large number of trials
(1250) were used to estimate the relative weights, the four methods
generated nearly identical weight estimates. When smaller numbers
of trials were used (112), the different methods generated patterns
of relative weights that were largely similar, and the patterns
deviated only modestly from the large-number solution. In terms
of efficiency, the error boundaries of the different methods were
nearly indistinguishable. All in all, the number of trials needed
to obtain statistically significant weights is sufficiently large
that there is no apparent advantage of using one method over the
others. [Work supported by NIH/NIDCD.]
|
|
<-- back to Recent Research
Recent Abstracts
Presentations & Posters
|
