2018 Conference on Cognitive Computational Neuroscience

5-8 September 2018    Philadelphia, Pennsylvania

Technical Program

Paper Detail

Paper: PS-1B.29
Session: Poster Session 1B
Location: Symphony/Overture
Session Time: Thursday, September 6, 18:45 - 20:45
Presentation Time:Thursday, September 6, 18:45 - 20:45
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: Data-driven methods reveal the generalizing mechanisms of speech processing in naturally varying soundscapes
Manuscript:  Click here to view manuscript
DOI: https://doi.org/10.32470/CCN.2018.1136-0
Authors: Moritz Boos, Applied Neurocognitive Psychology Lab, Germany; Jörg Lücke, Machine Learning Group, Germany; Jochem Rieger, Applied Neurocognitive Psychology Lab, Germany
Abstract: Humans rarely encounter speech without background noise. However, research on the cortical mechanisms of speech processing mostly focusses on individual speech features in isolation, which might not generalize to a more naturalistic environment. To examine the mechanisms of speech processing in natural soundscapes, we use unsupervised learning to infer spectro-temporal patterns that are adapted to the statistics of speech in noise. Using these patterns, we predict fMRI activity (n=20) evoked by a long auditory stimulus with voxel-wise encoding models and find a latent space of predicted brain activity that is shared between participants and represents the perceived noise level of the stimulus. Activity in the latent space forms two clusters, one representing stimuli with varying noise level related to the presence of simple, time-frequency separable patterns, and another consisting of stimuli with uniformly low perceived noise level. The cluster representing noisy stimuli explains variance in secondary auditory areas, through reduced activation for very noisy stimuli, while the cluster consisting of clear speech explains variance in both primary and secondary auditory areas. This shows how features adapted to speech in a natural soundscape relate to differences in the subjective percept of noise and the resulting dichotomy in brain activity.