Speech Perception Network Underpins the Generalization of Speech Category in Non-speech Contexts
Other conference paper

Extracting discrete meaningful speech units across various acoustic contexts is critical to speech perception. This process of categorization requires significant neural computations. In tone languages, in addition to segmental units, the brain needs to extract continuous time-varying pitch patterns and map onto linguistically-relevant tones. An area of intense debate in speech perception is the extent to which the computations underlying categorization across various contexts is mediated by a core auditory network and how this network relates to categorization decision. A primary goal of the current study is to understand the neural implementation of how the brain extracts and categorizes speech categories across a variety of contexts.
We synthesized non-speech iterated rippled noise (IRN) analogs of time-varying pitch patterns (i.e., T1: flat tone; T2: rising tone; T4: falling tone) using procedures described in Swaminathan et al. (2008). We parametrically manipulated the saliency of pitch with different iteration steps (i.e., 2, 4, 8, 16, and 32) to examine the extent to which perceptual salience impact the computations underlying categorization. We scanned native speakers of Mandarin Chinese participants (N = 31) using functional magnetic resonance imaging (fMRI), during which the participants were asked to listen to sounds (i.e., consisting of six conditions [IRN conditions with 2, 4, 8, 16, 32 iterations and speech condition] and three tone categories [T1, T2, T4] and two talkers [male and female]) and categorize them into one of the three categories based on their pitch patterns. We used a customized sparse-sampling scanning protocol (TR = 2.5 sec with 800-ms silence gap) to minimize the influence of scanner noise on speech perception, in which each sound was presented during the silent period. We used multivariate pattern classification (MVPC) to reveal the core neural representation of tone category, in which classifiers were trained with speech sound, and the model was generalized on non-speech sounds. Representational similarity analysis (RSA) was further used to reveal the tone category representation while controlled variance of other acoustic and non-acoustic factors. Finally, principal component analysis (PCA) and support vector regression were used to identify brain areas that are related to categorization decision.
Behaviorally, we did not find a significant difference between speech and non-speech conditions. The IRN2 (least perceptually salient) condition was responded significantly longer than other conditions. Univariate activation analysis revealed the classical fronto-tempo-parietal speech perception network, which is consistent with brain network derived from meta-analysis. MVPC showed that significantly higher-than-chance tone category decoding was identified in the bilateral middle portion of the superior temporal gyrus (LmSTG). RSA further demonstrated that categorical tone representation was identified in the LmSTG after controlling for all other acoustic and perceptual-based factors. PCA and RT-prediction analysis showed that the neural correlates of categorization decision processing are identified in the temporal and frontal regions, including the STG, LIFG, pre-CG, and SM.
Our MVPC and RSA results suggest that the neural representation of context-independent tone categories is evident in the middle portion of left STG. This core neural representation supports the extraction of tone categories across contexts that differ in surface acoustic properties as well as perceptual saliency. Categorization decision requires the involvement of a distributed brain network, including the tone representational STG regions, IFG and pre-central motor regions. Our findings altogether reveal the specific neural characteristics of frontotemporal regions that support context-independent speech category representation and categorization decision.
著者Gangyi Feng, Zhenzhong Gan, Danting Meng, Suiping Wang, Patrick Wong, Bharath Chandrasekaran
會議名稱The Organization for Human Brain Mapping Annual Meeting

上次更新時間 2019-20-06 於 15:33