Dr. Radhika Aravamudhan

Post-Doctoral Research Associate

As a graduate student, the first time when I was taught the “Motor Theory of Speech Perception”, I was thrilled by the idea that we had a special ability in encoding and decoding speech as a special form of auditory signal. I felt pride in possessing this special quality (One of those things like having an ‘opposing thumb’ I guess).

Soon after that, I was introduced to other theories of speech perception, where I learnt we encoded and decoded all auditory signals in a similar fashion, irrespective the type of auditory signal.

So speech is not perceived in a special mode?

This is the question that led me in this path of research in auditory perception. Here are some of the projects I am involved in.

Perceptual Overshoot in Speech and Nonspeech sounds: (Doctoral Dissertation-Kent State University)

The study addressed the differences in perception of speech and nonspeech signals by comparing the perceptual overshoot in synthetic vowels and sinewave acoustic replicas of the synthetic vowels. Lindblom and Studdert-Kennedy (1967) demonstrated that the perceptual boundary for steady state vowels and that for vowels in a CV context with F2 transition are different. This mechanism was called the perceptual overshoot, was thought to serve as a compensatory mechanism for the articulatory undershoot present during speech production. In experiment 1, elicitation of PO with speech and nonspeech signals was studied. Boundary estimates were obtained for four different continua for all 20 subjects: (1) steady-state synthetic vowels; (2) synthetic vowel transitions in /wV/ context; (3) sine wave analogs mimicking the steady-state vowels; and (4) sine wave analogs mimicking the transition continuum. Results replicated the findings of previous studies by demonstrating PO with synthetic vowels; however, PO was not elicited with the sinewave analogs. This led to a second experiment, where the role of training on the elicitation of PO with sinewave analogs was investigated. Subjects from the first experiment were divided into two groups. One group was trained to categorize the steady-state sinewave analogs based on their synthetic vowel boundary. Training occurred over 3–6 1-h sessions until subjects achieved 90%-correct identification. The control group received no training. Post-training boundary estimates were obtained for the four original continua from both groups. The results indicated that PO was now elicited with the sinewave analogs for the trained group, but not for the control group. The results support the hypothesis that processes involved in processing speech and nonspeech sounds maybe the same and that any differences obtained in the performance with speech and nonspeech may be due to lack of adequate exposure or training with the nonspeech sounds. Thus, the effects of context maybe universal to any target sound, as long as there are strong categories and are not dependent on phonetic relevancy of the category.

Training non-native speech contrasts:

- Dr.Radhika Aravamudhan & Dr. Anna M. Schmidt (Kent State University)

o Involves training English /v/ - /w/ contrast with Indian subjects

Parts of this work has been presented at the Acoustical Society of America meeting, 2001, Chicago, IL

Phonetic context effects in cochlear implant listeners:

- Dr. Radhika Aravamudhan & Dr. Andrew J Lotto

Perception of a phoneme changes as a function of the characteristics of preceding or following phonemes. These changes in percept are referred to as phonetic context effects. Experimentally this effect is measured in terms of a shift in phoneme identification (ID) boundary or a change in the percentage of phoneme ID that occurs when the same target stimuli are presented with different context sounds. Previous research (e.g., Lotto & Kluender, 1998) has demonstrated that many of these context effects are the result of interactions between the spectral patterns of the context and target sounds. It is likely that if the spectral representations of speech sounds are changed (as happens with cochlear implants), phonetic context effects will be affected.

Two types of contexts are used in this study: Spectral based and temporal based context.

Spectral-based context effects: Listeners identify a consonant or vowel that is preceded by phonemes that differ in their spectral pattern. NH listeners show a shift in responses that is predicted by the spectral relations of target and context (contrastive).

Temporal based context effect: Listeners identify a target consonant distinction that is temporal (e.g., /b/ vs. /w/) that is followed by a context vowel that varies in duration. NH listeners show a contrastive shift in target identification based on vowel duration.

Results: As predicted CI listeners showed normal temporal context effects but not spectral context

effects. This is probably because of the substantial deviation of spectral patterns but preserved duration patterns in CI input. The lack of normal spectral context effects may have practical implications for situations in which there is substantial coarticulation (e.g., non-laboratory speech) or talker variability (e.g., switching between multiple speakers).

Thus, in order to understand how these poor spectral representations in cochlear implants can affect the percept of speech sounds, further simulations experiments are being conducted with normal hearing listeners. Simulations are done using both Shannon Speech and CI-Simulation programs.

Parts of this work has been presented at the following meetings

· Acoustical Society of America meeting - 2004, SanDiego, CA

· 10th Symposium on Cochlear Implants in Children -2005, Dallas, TX

· Conference on Implantable Auditory Prosthesis -2005, Asilomar, CA