Project:
Vocal analytics: novel automated tools for decoding infant babbling patterns
Laboratories:
Roger Levy, Ph.D. and Elika Bergelson, Ph.D.
Biographical Information:
Marvin received his B.S. from Université Pierre et Marie Curie and M.S. from ENSIMAG in Mathematics and Computer Sciences. In 2023, he completed his Ph.D. in Cognitive Sciences at Meta AI Research and École Normale Supérieure, under the supervision of Alejandrina Cristia, Emmanuel Dupoux, and Hervé Bredin, where he developed artificial neural networks to analyze and simulate language acquisition in children.
Current Work:
What mechanisms shape the emergence of babbling patterns across languages? How early can we detect signs of atypical development in infants’ vocalizations? Studying these fundamental questions requires meticulous efforts to manually transcribe children’s linguistic productions, creating a significant bottleneck in understanding how babbling patterns emerge and evolve across linguistic and neurodevelopmental communities. A powerful key to addressing these questions may lie in automated analyses of naturalistic infant vocalizations using state-of-the-art speech processing tools.
In this project, I seek to develop novel computational tools to decode infant babbling patterns, examine their developmental trajectory, and investigate how they may differ between typical and atypical development. This investigation includes building a universal speech-sound recognizer for infant vocalizations, quantifying when and how babbling becomes language-specific, and identifying early vocal markers associated with typical and atypical development. Together, this work aims to accelerate research on early speech development through automated tools while providing insights that could enable earlier detection and more targeted interventions for developmental disorders affecting speech and language acquisition.
Publications:
Lavechin, M., Bousbib, R., Bredin, H., Dupoux, E., & Cristia, A. (2020). An open-source voice type classifier for child-centered daylong recordings. Interspeech.
Lavechin, M., Métais, M., Titeux, H., Boissonnet, A., Copet, J., Rivière, M., … & Bredin, H. (2023, December). Brouhaha: multi-task training for voice activity detection, speech-to-noise ratio, and C50 room acoustics estimation. In 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) (pp. 1-7). IEEE.
Lavechin, M., De Seyssel, M., Gautheron, L., Dupoux, E., & Cristia, A. (2022). Reverse engineering language acquisition with child-centered long-form recordings. Annual Review of Linguistics, 8(1), 389-407.
Lavechin, M., de Seyssel, M., Métais, M., Metze, F., Mohamed, A., Bredin, H., … & Cristia, A. (2024). Modeling early phonetic acquisition from child-centered audio data. Cognition, 245, 105734.
Keywords:
machine learning, speech processing, typical and atypical language development, autism spectrum disorders