Guillem Bonafos
- Sensory Neuro-Ethology (CRNL-ENES)
- Laboratoire Hubert Curien (LabHC)
-
guillem-bonafos
-
0000-0002-7111-0288
-
285520822
-
bonafos_g_1
- ResearcherID : ABF-9875-2021
Présentation
My research is driven by a fundamental question: how does language emerge? I investigate this question by studying children’s vocalizations during the pre-linguistic stage, viewing them as a window into the mechanisms that precede and shape language acquisition. In particular, I examine the information structure of early vocal behavior and the extent to which these signals can predict subsequent developmental trajectories. Methodologically, my work combines machine learning and topological data analysis to uncover the structure of high-dimensional vocalization data. I aim to learn representations that are both maximally informative and intrinsically interpretable, allowing links to be drawn between computational structure and developmental processes. At the same time, the developmental dynamics of vocalizations provide a principled framework for probing, constraining, and ultimately disentangling these learned representational spaces.
Publications
Publications
|
|
Topological data analysis of human vowels: Persistent homologies across representation spacesSpeech Communication, 2026, 178, pp.103363. ⟨10.1016/j.specom.2026.103363⟩ |
|
|
Premature baby cries remain different from those of full-term babiesJournal of the Acoustical Society of America, 2025, 157 (6), pp.4175 - 4183. ⟨10.1121/10.0036842⟩ |
|
|
Detecting non-adjacent dependencies is the exception rather than the rulePLoS ONE, 2022, 17 (7), pp.e0270580. ⟨10.1371/journal.pone.0270580⟩ |
|
|
Learning Higher‐Order Transitional Probabilities in Nonhuman PrimatesCognitive Science, 2022, 46 (4), ⟨10.1111/cogs.13121⟩ |
|
|
Detection of regularities in a random environmentQuarterly Journal of Experimental Psychology, 2020, pp.174702182094135. ⟨10.1177/1747021820941356⟩ |
|
|
Speech transformer models for extracting information from baby criesWorkshop on Child Computer Interaction - WOCCI 2025, Aug 2025, Nijmegen, Netherlands. pp.11 - 15, ⟨10.21437/wocci.2025-3⟩ |
|
|
Dirichlet process mixture model based on topologically augmented signal representation for clustering infant vocalizationsInterspeech 2024, ISCA, 2024, Kos, Greece. pp.3605-3609, ⟨10.21437/Interspeech.2024-394⟩ |
|
|
Application to long recordings of babies: Automatic extraction of infant vocalizations from one year home audio recordingsWorkshop on Infant Language Development (WILD-2024), Jun 2024, Lisbonne, Portugal |
|
|
Detection and classification of vocal productions in large scale audio recordings2023 |
|
|
Classification non-supervisée des productions vocales chez le bébé humain entre 0 et 12 moisStatistiques [math.ST]. Aix-marseille University, 2023. Français. ⟨NNT : 2023AIXM0487⟩ |