Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Rate-normalized Encoding of Vowel Duration during Speech Processing
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Mara Wolter1,2,3, Andrey Zyryanov1,2,3, Yulia Oganian1,2,3; 1Centre for Integrative Neuroscience, University Medical Center Tübingen, Germany, 2Graduate Training Centre of Neuroscience, University of Tübingen, Germany, 3International Max Planck Research School for The Mechanisms of Mental Function and Dysfunction, Tübingen, Germany
How fast or slow someone talks differs between speakers, and listeners interpret durational cues in speech relative to this contextual speech rate. Consequently, listeners perceive a fixed duration as longer in a fast-rate context than in a slow-rate context. Prior work has shown that auditory cortices track contextual speech rate and encode durational information, yet how these cues are integrated towards perception remains unclear. A rate-normalized encoding account posits that contextual speech rate directly shapes sensory processing, such that fast-rate contexts would shift neural activity toward representations of longer durations. Under a rate-independent encoding account, sensory representations would independently encode absolute durations and the contextual speech rate, with speech-rate normalization occurring at later stages. To contrast these hypotheses, we recorded MEG, while participants (14m/15f/1d) judged whether German vowel duration minimal word pairs (satt–Saat, Bann–Bahn, Rum–Ruhm) contained a long or short vowel. Word pairs were presented along seven-step vowel-duration continua, each preceded by a five-syllable sentence spoken at either a fast (7 syl/s) or slow (2.5 syl/s) rate. As expected, we observed a behavioral rate normalization effect: The fast-rate context increased long-vowel responses by 13.6% (p<0.001) compared to the slow-rate context. Encoding modeling of source-reconstructed MEG data showed that rate modulated neural activity throughout the vowel offset response in bilateral auditory, sensorimotor, and inferior frontal cortices, peaking ~130 ms post vowel offset. Vowel duration effects emerged in right posterior perisylvian areas from ~70 ms and extended between ~150-400 ms in the right auditory cortex. Comparison of neural patterns for rate and duration representations using cosine similarity showed that duration encoding was embedded in rate encoding patterns and aligned to them throughout encoding (p<0.001), consistent with a rate-normalized encoding account. Spatially, alignment was strongest in bilateral auditory temporal, sensorimotor, and right inferior frontal regions. We then asked whether fast-rate context shifts neural activity toward patterns associated with longer vowel durations, as expected for rate-normalized representations. For this, we projected individual neural activity patterns onto a representational subspace capturing vowel duration encoding. Analyses of rate effects along this space showed that fast-rate context shifted neural responses toward the long-vowel pattern (p<0.001). Critically, trial-by-trial expression of this pattern predicted participants’ behavior beyond contextual rate and absolute vowel duration (p<0.001), linking these contextualized neural representations directly to perception. Lastly, we asked whether contextual-rate representations remain stable throughout target processing or are dynamically updated following the target word. Neural responses discriminated rate contexts both in the pre-target silence period and following target word offset. However, relative to the pre-target pattern, after vowel offset the rate effect was inverted, starting about ~330 ms post offset. This suggests that the contextual rate estimate is updated to incorporate the speech rate of the target word. This is consistent with a dynamical update of the rate estimate in auditory and sensorimotor cortices. Overall, our results suggest that auditory speech cortex concurrently tracks and integrates individual speech input and context statistics, forming the basis for contextually informed speech percepts.
Topic Areas: Speech Perception,