Search Abstracts | Symposia | Slide Sessions | Poster Sessions
EEG evidence for acoustic cue weights in speech perception
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Chiara Repetti-Ludlow1, Joseph Toscano2, Christina Bjorndahl1, Barbara Shinn-Cunningham1; 1Carnegie Mellon University, 2Villanova University
Introduction. A core principle in phonetics is that phonemes are identified based on acoustic cues. For example, previous research has shown that voice onset time (VOT) is the primary cue for stop voicing in English, allowing listeners to distinguish perceptually between [b] and [p] in words like “beach” and “peach.” EEG studies support this finding, demonstrating that in mapping an acoustic signal to phonological categories, the P3 component is sensitive to VOT, such that P3 amplitude is greater when VOT is more prototypically “voiced” or “voiceless” as opposed to ambiguous. However, listeners rely on multiple cues with different perceptual weights. For example, in English stop voicing, the fundamental frequency (f0) plays a secondary role in phoneme identification, co-occurring with VOT and compensating for masked primary cues. While secondary acoustic cues are central in speech perception, it is still unclear whether they are processed similarly to primary acoustic cues, and whether neurological data reflects phonetic theories that cues exist in a weighted relationship. The present study addresses this gap by considering (1) whether P3 is sensitive to secondary acoustic cues such as f0, and (2) whether P3 can be used as an assay of acoustic cue weights. Methods. An EEG study was designed using the oddball paradigm with stimuli varying along two stop voicing continua: beach~peach [b~p] and dear~tear [d~t]. Each continuum was manipulated in Praat for VOT (5 steps) and post-stop f0 (3 steps), yielding 15 stimuli per continuum. In each of the four blocks one of the words is the target word, and in each trial, participants hear a stimulus and press one button if it is the target and another if it is not. Each stimulus is presented 10 times in each block for a total of 1200 trials. ERP analyses were carried out with statistical validation via linear mixed-effects models. Data collection is ongoing (n=12, target=30). Results. Behavioral results show that both VOT and f0 contribute to the perception of a sound as voiced or voiceless, with a logistic mixed-effects model showing significant results for both VOT (β=0.83, p=<0.01) and f0 (β=0.37, p=<0.01), but a greater effect for VOT. The ERP analysis confirms that more prototypically voiced and voiceless VOTs elicit greater P3 amplitudes, and demonstrate the novel finding that more prototypically voiced and voiceless f0s also elicit greater P3 amplitudes. Finally, when we compare VOT and f0, results suggest that cue weighting is playing a role, with prototypical VOTs showing a larger P3 than prototypical f0s, and mid-continuum/ambiguous cues eliciting a smaller P3. Conclusion. Preliminary results paint a picture of the neural processing of phonemes that is largely in line with phonetic theory. Behavioral and ERP data demonstrate that both VOT and f0 contribute to the perception of a phoneme as voiced or voiceless, but VOT plays a stronger role. When considering the interaction of the two cues, we see evidence for cue weights, suggesting that EEG data could be a tool for determining acoustic cue weights in future research.
Topic Areas: Speech Perception, Phonology