Grouping Cues Research Articles

Evaluation of a representational model for auditory scene analysis

Physiological data suggest that a two-dimensional signal representation of amplitude-modulation frequency against center frequency is extracted in the central nucleus of the inferior colliculus [C. E. Schreiner and G. Langner, J. Neurophysiol. 60, 1823–1840 (1988)]. The representation groups signals with common harmonics, and can easily be adapted to also enhance common onsets. This makes it an attractive basis for a representational model of auditory scene analyis. The map is used as a front end to a neural network pattern matching stage for vowel recognition. The model is tested against human performance in a concurrent vowel recognition task [P. F. Assmann and Q. Summerfield, J. Acoust. Soc. Am. 88, 680–697 (1990)]. It predicts human performance for a F0 grouping task, even for vowels with up to a 12-dB intensity difference, well. Another grouping cue, demonstrated in abstract sounds, is common onset. The double-vowel task was modified to include onset asynchrony. A model based on the change in the AM-map representation between successive frames predicts human performance well. It is possible to directly drive a pattern matching stage with the AM map representation removing the need for an explicit grouping processes. The model is inherently robust when driven by stimuli that violate grouping cues.

Generalizing solution procedures learned from examples.

Three experiments tested the hypothesis that when learners are led to group steps from example solutions, they will be more likely to learn subgoals that can be transferred to novel problems, thereby improving problem solving. The results from each experiment suggest that a label serves as a cue for grouping by demonstrating that a relatively meaningless label in an example solution is as effective as a meaningful label in helping learners transfer to novel problems. Experiments 2 and 3 provide converging evidence for subgoal learning by demonstrating that participants studying example solutions with a label were more likely to segment the solution as a function of the label, to mention the corresponding subgoal in their self-explanations while studying the examples, and to mention the subgoal in their descriptions of how to solve problems after they studied the examples.

Perceptual organization in a comodulation masking release interference paradigm: exploring the role of amplitude modulation, frequency modulation, and harmonicity.

Two areas of interest within auditory scene analysis are the identification of cues which contribute to auditory grouping and/or segregation, and the methodology employed to measure such effects. The present study explored the roles of three purported grouping cues--correlated AM, correlated FM, and harmonicity--using a comodulation masking release interference paradigm. This method was reasoned to be free of alterations in the subject's mode of listening (analytic versus synthetic). The results suggested that correlated AM was a viable grouping cue. For some listeners, the presence of FM also appeared to provide a grouping cue, but, generally, no effect of coherence of FM was observed. No effect of harmonicity was observed in this paradigm.

Perceptual grouping by similarity and proximity: Experimental results can be predicted by intensity autocorrelations

A model for perceptual grouping based on measurements of spatial correlations is proposed and tested. Gestalt-like grouping experiments were performed to study and quantify the effect of element similarity (shape, luminance) and proximity. Observers reported the horizontal or vertical organization of stimuli with proximity and similarity providing conflicting grouping cues. Proximity grouping was found to be perceived much faster than similarity grouping. However, with increasing processing time, similarity was found to dominate grouping. The experimental results can be accounted for by assuming a process that compares horizontal and vertical intensity autocorrelations. The model suggests that correlations are measured across a limited spatial range, and that this range increases with processing time.

Comparison of the effect of onset asynchrony on auditory grouping in pitch matching and vowel identification.

Previous experiments have shown that when a slightly mistuned harmonic of a complex tone starts more than about 80 msec before the remaining components, it makes a reduced contribution to the pitch of the complex. This contribution decreases to zero by about 300-msec onset asynchrony. In vowel perception, however, analogous experiments have shown that a much shorter asynchrony (around 40 msec) is enough to ensure that a component does not influence a vowel's phonemic category. The three experiments reported here demonstrate that this difference in the utility of onset time as a grouping cue does not arise because of differences in stimulus structure, but rather is due to the perceptual task. They show that the onset asynchrony needed in a pitch-matching experiment to remove the contribution that a mistuned component makes to the pitch of a vowel is the same as that needed to remove the contribution to the pitch of a flat-spectrum complex tone. They further show that a much smaller onset asynchrony is needed to perceptually remove the same harmonic from a vowel for the calculation of vowel quality. The implication of this result for models of auditory grouping is discussed.

Positive Transfer across Grouping Cue Modalities in Rat Serial Learning

In Experiment 1, three groups of rats ( n = 4) received four successive repetitions daily of the three-trial series of reinforced (R) and nonreinforced (N) trials RNR. In Group B, a change in runway brightness (/) served as a grouping cue to signal the start of each series repetition, e.g., RNR/RNR/RNR/RNR. In Group T, a relatively long intertrial interval served as a grouping cue, e.g., RNR–––RNR–––RNR–––RNR. Group C was trained in Phase 1 with no grouping cue, e.g., RNRRNRRNRRNR. Groups B and T were superior to Group C in learning to anticipate the events of the RNR series, a result consistent with the hypothesis that those groups would organize the 12-item list into groups of three trials presented recursively. Following Phase 1, groups were shifted from temporal grouping to brightness grouping and vice-versa and to a tactile grouping cue condition. Except for shifts from temporal to brightness grouping, correct responding survived cross-modal shifts in grouping cues. Experiment 2 employed a subtly modified series of reinforcement events and showed that positive transfer could also be obtained in subjects shifted from temporal to brightness grouping. The results support the view that grouping cues employed in rat serial learning investigations are not discriminative cues for particular groupings of reinforcement events.

Perceptual separation of simultaneous vowels: Within and across-formant grouping by F

Six experiments explored why the identification of the two members of a pair of diotic, simultaneous, steady-state vowels improves with a difference in fundamental frequency (delta F0). Experiment 1 confirmed earlier reports that a delta F0 improves identification of 200-ms but not 50-ms duration "double vowels"; identification improves up to 1 semitone delta F0 and then asymptotes. In such stimuli, all the formants of a given vowel are excited by the same F0, providing listeners with a potential grouping cue. Subsequent experiments asked whether the improvement in identification with delta F0 for the longer vowels was due to listeners using the consistent F0 within each vowel of a pair to group formants appropriately. Individual vowels were synthesized with a different F0 in the region of the first formant peak from in the region of the higher formant peaks. Such vowels were then paired so that the first formant of one vowel bore the same F0 as the higher formants of the other vowel. These across-formant inconsistencies in F0 did not substantially reduce the previous improvement in identification rates with increasing delta F0's of up to 4 semitones (experiment 2). The subjects' improvement with increasing delta F0 in the inconsistent condition was not produced by identifying vowels on the basis of information in the first-formant or higher-formant regions alone, since stimuli which contained either of these regions in isolation were difficult for subjects to identify. In addition, the inconsistent condition did produce poorer identification for larger delta F0's (experiment 3). The improvement in identification with delta F0 found for the inconsistent stimuli persisted when the delta F0 between vowel pairs was confined to the first formant region (experiment 4) but not when it was confined to the higher formants (experiment 6). The results replicate at different overall presentation levels (experiment 5). The experiments show that at small delta F0's only the first-formant region contributes to improvements in identification accuracy, whereas with larger delta F0's the higher formant region may also contribute. This difference may be related to other results that demonstrate the superiority of resolved rather than unresolved harmonics in coding pitch.

Extinction of taste aversion does not eliminate taste-mediated aversion to visual cues: Replicating Lett (1984)

In this experiment, we evaluated whether extinction of taste aversion following conditioning with a compound CS of visual and taste cues would attenuate taste-mediated aversion to visual cues. Sixty chicks (Gallus domesticus) received a pairing of a compound CS, red-colored water or red-colored vinegar, with a toxin and then extinction procedures with clear vinegar. Testing with red water showed reliable evidence of potentiation to visual cues in the red vinegar conditioning groups, as measured by latency to start drinking and intake performance. Taste-aversion extinction groups (n = 30), tested with clear vinegar, showed no reliable evidence of taste aversion. The results contradicted the within-compound hypothesis of taste-mediated potentiation and replicated Lett (1984).

The contribution of onset, offset, and comodulation to the grouping of noise bands

Auditory grouping uses two difference kinds of cues; ‘‘vertical’’ curs act on small temporal sections and are based on similarity principles (frequency, harmonicity, onset/offset, modulation), whereas ‘‘horizontal’’ (sequential) cues depend on the history of the stimulus. The present study addresses the relative importance of several vertical grouping cues. Twenty naive subjects were presented with auditory scenes of one to five elements. The elements were noise bands with a width of 50 Hz and a duration of 0.12 to 0.36 s. The center frequencies, the onsets and the offsets were randomized; in half of the scenes the noise bands were comodulated. The subjects were asked how many events they had heard. The number of perceived auditory events was on the average smaller than the number of presented noise bands. This could be shown to depend mostly on onset synchrony, to a slightly smaller degree on offset synchrony, and only very little on comodulation.

Scaling the perceptual clarity of the presence or absence of target tones embedded in modulated noise maskers

Perceptual judgments of the presence or absence of tones embedded in modulated noise maskers were obtained using a six-point rating scale (1=clearly absent, 6=clearly present). The maskers comprised four simultaneous noise bands centered at 500, 1000, 1500, and 2000 Hz, with bandwidths of all bands either 12.5, 25, 50, or 100 Hz. The bands were derived either from the same noise source by filtering and transposition, or from different sources, and were subsequently multiplied by same or different (phase shifted) modulators. A sine tone at the center frequency of one of the noise bands was used as the signal. A single noise interval was preceded by a cue tone that was either present or absent in the noise at a fixed S/N≊−13 dB SPL. The cue and noise intervals were both 400 ms long, and separated by 1 s of silence. Rating scores from ten normally hearing listeners were generally higher for tones embedded in maskers with common envelopes, indicating less masking and improved perceptual clarity of the tones. Common origin of noise source did not lead to improved audibility of the tones. There was no significant effect of tone frequency. However, bandwidth affected judgments significantly. For the 12.5- and 25-Hz bandwidths, the difference between modulation conditions was reduced, while false alarm rate increased substantially. These narrow-band noises apparently sounded tonal even in the absence of the signal, thereby receiving higher ratings. For the 50- and 100-Hz-wide maskers, the effect of modulation pattern was enhanced, and there were fewer false alarms. Similarity of amplitude envelopes within a masker has been invoked as a simultaneous grouping cue facilitating segregation of a concurrent signal. It seems that sequential grouping factors such as similarity of perceptual attributes (e.g., pitch or timbre) of the cue and masker also play a role in signal detection. [Work supported by NSERC.]

The effect of amplitude comodulation on auditory object formation in sentence perception.

To comprehend speech in most environments, listeners must combine some but not all sounds from across a wide range of frequencies. Three experiments were conducted to examine the role of amplitude comodulation in performing an essential part of this function: the grouping together of the simultaneous components of a speech signal. Each of the experiments used time-varying sinusoidal (TVS) sentences (Remez, Rubin, Pisoni, & Carrell, 1981) as base stimuli because their component tones are acoustically unrelated. The independence of the three tones reduced the number of confounding grouping cues available compared with those found in natural or computer-synthesized speech (e.g., fundamental frequency and simultaneity of harmonic onset). In each of the experiments, the TVS base stimuli were amplitude modulated to determine whether this modulation would lead to appropriate grouping of the three tones as reflected by sentence intelligibility. Experiment 1 demonstrated that amplitude comodulation at 100 Hz did improve the intelligibility of TVS sentences. Experiment 2 showed that the component tones of a TVS sentence must be comodulated (as opposed to independently modulated) for improvements in intelligibility to be found. Experiment 3 showed that the comodulation rates that led to intelligibility improvements were consistent with the effective rates found in experiments that examined the grouping of complex nonspeech sounds by common temporal envelopes (e.g., comodulation masking release; Hall, Haggard, & Fernandes, 1984). The results of these experiments support the claim that certain basic temporal-envelope processing capabilities of the human auditory system contribute to the perception of fluent speech.

Hierarchical organization of series information by rats: Series chunks and list chunks

In each of three runway experiments, rats received 24 trials per day organized by temporal grouping into two daily repetitions of two discrete six-item series (Series A and B). In turn, instructions to organize both series into two discrete lists (Lists 1 and 2) were presented in the form of brightness (Experiment 1) or tactile (Experiments 2 and 3) grouping cues. List 2 of Series A consisted of the three trial outcomes reinforcement (R), nonreinforcement (N), reinforcement (R) in each experiment; List 2 of Series B was always RNN. In each experiment, rats learned to anticipate all of the events of List 2, running fast to R and slow to N. Of particular interest was the development of differential responding to the terminal R and N events, behavior that could not have been regulated by intralist information as those trials were preceded by RN in both series. The data suggest List 2 was represented as a functional unit, or list chunk, its elements anticipated on the basis of a representation of one or more items from List 1. At a minimum, the List 1 items that signaled List 2 consisted of the last item in List 1 (Experiment 1), the first item in List 1 (Experiment 2), and a representation of List 1 items 1 and 2 that maintained information about their order. Transfer tests accompanying Experiment 1 yielded an important novel finding that grouping cues are functionally independent of the items they serve to organize.

The effect of rate of amplitude comodulation on sentence perception.

Previous studies of the perception of time-varying sinusoidal (TVS) sentences suggested that amplitude comodulation may serve as a grouping cue in speech perception [T. D. Carrell, J. Acoust. Soc. Am. Suppl. 1 86, S102 (1989)]. The purpose of the present investigation was to further explore that idea and to evaluate whether the effects observed in TVS studies are related to other temporally based phenomena such as comodulation masking release (CMR). If a common process underlies the perception of amplitude-modulated speech and nonspeech stimuli, then grouping effects should occur in speech perception in a manner similar to that observed in nonspeech studies when the amplitude modulation characteristics are similar. Examined in the present study was the effect of amplitude modulation rate on intelligibility of TVS sentences as compared with rate effects in CMR phenomena. Amplitude modulation rates were 5, 10, 20, 50, 100, and 200 Hz. Although the results suggested that comodulation is important in speech perception, effective rates may be different from those observed in nonspeech tasks. These findings are important for our understanding of auditory object formation of both speech and nonspeech stimuli.

The relation between auditory grouping and comodulation masking release.

The feature of coherent amplitude modulation across frequency, which gives rise to the phenomenon of comodulation masking release (CMR), is also considered to be a cue for auditory grouping. This commonality raises the question of whether it is a prerequisite for CMR that the correlated noise bands comprising the masker be grouped together. The present study measured signal threshold in the presence of correlated bands of noise under conditions designed to discourage the fusion of the noise bands. In the first set of conditions, the onset/offset synchrony between the noise band centered on the signal and the comodulating flanking bands was manipulated. The results indicated that the CMR measured when the on-signal band and flanking bands were gated synchronously was largely eliminated with onset/offset asynchronies of 50–100 ms (on-signal band beginning before and ending after the flanking bands). In the second set of conditions, signal threshold in the on-signal band was measured in the presence of a single flanking band, either with or without a background harmonic complex that included in its harmonic series the center frequency of the flanking band, but not the on-signal band. No consistent effect of harmonicity was observed. [Work supported by AFOSR.]

Effects of age at castration on testosterone induced aggression-promoting cues in groups of male mice

Groups of five male albino mice were set up in observation cages so that each contained one individual from each of the following five categories: trained fighters (T.F.); mice castrated pre-pubertally and implanted with testosterone in adulthood (P.I.); mice castrated pre-pubertally but sham-implanted in adulthood (P.S.); mice castrated in adulthood and implanted (A.I.) and mice castrated in adulthood but sham-implanted (A.S.). T.F's. became dominant to all these mice but did not distribute their attacks uniformly among the four categories of opponent. A.I. mice were attacked significantly more than P.I. mice which in turn were attacked significantly more than either P.S. or A.S. mice. It was concluded that the age at castration affected the response of mice to testosterone in terms of the production of aggression-eliciting olfactory cues. This effect seems likely to be a consequence of the duration of time between castration and hormonal treatment rather than an effect of age per se.