Carlyon

Slide1: 

The Continuity Illusion, Frequency Modulation, Vowel Perception, and Elephant Noise Bob Carlyon John Deeks Dennis Norris Sally Butterfield Christophe Micheyl Rhodri Cusack MRC Cognition and Brain Sciences Unit, Cambridge, England.

Overview: 

Overview Introduction to the continuity illusion (CI) Frequency modulation (FM - “vibrato”): Feature detectors? Encoding of FM phase Experiments on CI and FM Experiments on CI and vowel identification “Elephant noise” Themes: Can CI tell us about other processes? Constraint on where and when CI occurs Development of performance measures

The continuity illusion (“CI”): 

This “continuity illusion” only occurs if the level and frequency content of the inducing sound means that it would have masked the interrupted sound if it had remained on all the time. The continuity illusion (“CI”) When a sound is turned on and off briefly, and the silent gap is filled by an “inducing sound”, it can be heard as continuous. Allows us to perceive a sound as “whole” even when portions of it are masked.

CI: Constraints on neural bases?: 

CI: Constraints on neural bases? Sometimes described in neural terms, but non-specific e.g. “in every case, the neural stimulation that is characteristic of the restored sound can be found as part of the stimulation provided by the interrupting sound” (Bregman, Auditory Scene Analysis, p. 372). In periphery, the pattern of phase locking in neurons to a pure tone, or a glide, will differ vastly from that produced by a noise. So, if true, operates at stage where fine temporal information is discarded

CI: Constraints on neural bases?: 

CI: Constraints on neural bases? Sometimes described in neural terms, but non-specific e.g. “in every case, the neural stimulation that is characteristic of the restored sound can be found as part of the stimulation provided by the interrupting sound” (Bregman, Auditory Scene Analysis, p. 372). In periphery, the pattern of phase locking in neurons to a pure tone, or a glide, will differ vastly from that produced by a noise. So, if true, operates at stage where fine temporal information is discarded Subjective reports don’t distinguish between classes of explanation Are there feature detectors e.g. for glides, periodic FM, tonality?

Detectors for FM? (Cusack & Carlyon, 2000): 

Detectors for FM? (Cusack & Carlyon, 2000) An example trial. Condition 1, target present Condition 1 : odd warble Condition 2 : odd tone Suprathreshold; vary number of distractors

Results: 

Results

Frequency modulation (FM) phase and the Continuity illusion (“CI”).: 

Frequency modulation (FM) phase and the Continuity illusion (“CI”). Some evidence for amplitude modulation (AM) detectors, tuned to different rates. Less for FM, and listeners can’t compare pattern (phase) of FM between simultaneous tones. Plomp (1982): if a noise fills a silent gap in an FM tone, listeners hear the modulation continue. We asked whether the phase of the FM is preserved during the illusion

Detecting Physical FM Phase Reversals: 

Detecting Physical FM Phase Reversals If we’re going to investigate sensitivity to illusory phase reversals, need to study physical reversals Used an 800-ms 1-kHz sinusoid, modulated by +/- 10% 2-interval task: in signal, FM phase reversal mid-way through: Randomised starting FM phase: 0º or 180º Result: When FM rate=5 Hz, performance >90% correct

FM and CI: Experiment 1: 

FM and CI: Experiment 1 5 normal listeners 2-interval forced-choice 1 kHz carrier, 5 Hz FM +/- 10%, 60 dB SPL. 4 conditions: IN ALL CONDITIONS: Randomise starting phase Randomise duration Duration = 400 ms before and after reversal Noise=800-1200 Hz, 78 dB SPL.

FM and CI: Expt 1 Results: 

FM and CI: Expt 1 Results Subjects hear FM as continuous, but can’t hear phase reversal! Prediction: If FM phase is preserved in illusion, should do well in “Noise” condition

FM & CI: Experiment 2: 

FM & CI: Experiment 2 In experiment 1, noise failed to improve performance. How do we know that it induced continuity? Need to design an experiment where the perceived continuity affects performance. Kept duration fixed, and instructed subjects on how to do task in the gap condition: Can compare FM phase at start of each burst ..but not if the tone is heard as uninterrupted! ..or maybe even if joined by steady portion:

FM & CI, Experiment 2: Conditions: 

FM & CI, Experiment 2: Conditions Included continuous (physical reversal) condition, plus GAP, NOISE, and STEADY shown in previous slide. If performance is poor in NOISE condition, maybe due to masking tone onsets/offsets, or to a general distraction effect. Hence 2 new conditions, both with noise that does NOT induce illusion:

FM & CI, Experiment 2: Results: 

FM & CI, Experiment 2: Results Predictions are correct Only slightly impaired by off-freq or gapped noise Prediction: high scores with silent gap and when physically continuous; poor with illusory continuity Also poor when gap filled with steady tone Subjects can compare FM across but not within sounds?

FM & CI: a new illusion: 

FM & CI: a new illusion In this illusion, we replace the bandpass noise with another FM tone, which is same as inducee but louder & modulated at a lower rate (5 Hz): It sounds as if the faster FM rate is continuing behind the slower rate - even though there is no physical stimulus in which you can hear 5 and 15 Hz FM at the same time! -further evidence that FM is being encoded explicitly in the brain?

FM & CI, Experiment 2: Conclusions: 

FM & CI, Experiment 2: Conclusions Although subjects hear FM continue through a noise, no information on phase is preserved. This can lead to a paradoxical percept (hear FM, but don’t notice what would be an obvious phase reversal). Consistent with some neurons signalling existence of FM, but not encoding phase. Performance measure of continuity illusion: makes subjects worse. Subjects don’t have conscious access to “pre-illusion” stage of processing

Slide18: 

The Continuity Illusion and Vowel Identification

CI and vowel identification: Questions: 

CI and vowel identification: Questions “er” Vowel identification usually requires combination of simultaneous formants: Can we use the CI to trick the brain into combining information from non-simultaneous formants? If so, we can conclude that neural mechanisms responsible for CI “feed into” those involved in vowel identification “ee”

CI and vowel identification: 

CI and vowel identification Main comparison: identification performance in 3 conditions:

CI and vowel identification: Stimuli: 

CI and vowel identification: Stimuli Four 2-formant vowels: N.B. Only two different F1s Need to combine information across bands to identify vowels reliably (c.f. Bashford & Warren, 1987)

Preliminary experiment: 

Preliminary experiment Formant duration = noise duration = 100 or 200 ms. Total duration = 1.2 sec. Did the same with a 624-4000 Hz noise, and found level which induced continuity for all F2 but no F1s

CI and vowel identification: Method: 

CI and vowel identification: Method 12 subjects, all received training in a condition with each of 4 vowels presented continuously for 1.2 sec. This was the only condition in which feedback was given Six other conditions run in counterbalanced order. For each condition, duration of noise and formant segments=100 or 200 ms. Duration had no effect, and will be ignored. Calculated d’ for each subject, vowel, and condition. Averaged across subjects and vowels.

CI and vowel identification: conditions: 

CI and vowel identification: conditions

CI and vowel identification: Results: 

CI and vowel identification: Results Attending to F1 alone => d’=0.7

CI and vowel identification: Conclusions: 

CI and vowel identification: Conclusions Vowel identification improved by illusory continuity Not as good as physically simultaneous (noise partially masks formants?) Performance measure of continuity illusion (c.f. Plack and White, 2000); makes subjects better. Neural mechanisms for integrating information across frequency, for vowel identification, receive input from those involved in CI.

Retrospective continuity illusion: 

Retrospective continuity illusion Does the presence of the wideband noise (WBN, “body”) increase the perceived duration of the subsequent narrowband noise (NBN, “trunk”)?

Retrospective continuity illusion: 

Retrospective continuity illusion Measured perceived duration of NBN, compared to subsequent isolated NBN: At longest durations, does effect asymptote or decrease? (Rauschecker: sub-threshold depolarisation of BW-tuned neurons?)

Effect of WBN duration: 

Effect of WBN duration 8 subjects adjusted duration of isolated NBN until duration heard as equal to 1st NBN . 2 conditions: WBN=1000-4000 Hz WBN=5000-8000 Hz (controls for bias etc). Variable across subjects

Retrospective continuity illusion: 

Retrospective continuity illusion The WBN increases perceived duration of NBN by up to 80 ms Consistent with NBN being heard as starting earlier. Although consistent with auditory cortex, effect of WBN duration not consistent with sub-threshold depolarisation

Constraints on level of processing at which CI occurs: 

Constraints on level of processing at which CI occurs Receives input from binaural mechanisms (Kashino and Warren, 1996) Feeds into: Pitch perception (Plack & White, 2000) Vowel identification Acts retrospectively for about 80 ms (?) Preserves FM, but not phase information Can be measured objectively, and can improve or impair performance. Impairment=> no conscious access to “pre-illusion” processing.

Conclusions: FM and Vowel Identification: 

Conclusions: FM and Vowel Identification FM. Some evidence that there are feature detectors, and that phase may not be explicitly encoded.(Consistent with other findings). Vowel identification: Across-frequency process receives input from mechanisms underlying CI

Slide33: 

Post-doc job in Cambridge 2 years from June 2001 “Dynamic processing of frequency changes” Working with Bob Carlyon & Brian Moore Lovely location Nice people Contact: bob.carlyon@mrc-cbu.cam.ac.uk Free T-shirt

Physical FM Phase Reversals: Results: 

Physical FM Phase Reversals: Results Performance declines above 5 Hz Poor performance at high FM rates hints that FM phase may not be accurately encoded. Maybe frequency averaged over some temporal window? Mean of 5 subjects

Retrospective continuity illusion: 

Retrospective continuity illusion Preliminary experiments: perceived duration is increased. NBN duration of 300 ms gives large effect What happens as you increase WBN duration? At longest durations, does effect asymptote or decrease? (Rauschecker: sub-threshold depolarisation of BW-tuned neurons?)

Our approach:: 

Our approach: Try to constrain the level of processing at which illusion occurs Relationship to: Frequency Modulation Vowel identification Development of performance measures 3rd study looks at “retrospective” aspect of illusion: