iiinova — where your digital self keeps evolving

👏 Congratulations to the whole team, and especially to Marc-Antoine Georges and Marvin Lavechin! This work has been conducted at GIPSA-lab (@cnrs.fr/ Grenoble Alpes University) and is supported by the MIAI Cluster AI institute.

010

🧠 Why this matters? These works contribute to the development of computational models that learn acoustic, articulatory, and linguistic structure with minimal supervision and can be used to study the mechanisms underlying speech acquisition in children.

From perception to production: how acoustic invariance facilitates articulatory learning in a self-supervised vocal imitation model

2) From perception to production How acoustic invariance facilitates articulatory learning in a self-supervised vocal imitation model 📍 Gather Session 1 — 5 Nov 2025 @ 08:00 (online) 🔗 Full text: arxiv.org/abs/2509.05849 🎧 Demo&code: marvinlvn.github.io/projects/fro...

arxiv.org

Human infants face a formidable challenge in speech acquisition: mapping extremely variable acoustic inputs into appropriate articulatory movements without explicit instruction. We present a computati...

The 2025 Conference on Empirical Methods in Natural Language Processing

🎙️ Check out our two papers presented today at #EMNLP2025! 1. Decode, Move and Speak! Self-supervised vocal imitation linking acoustics, articulation & discrete speech units (originally published in Comp. Ling.) 📍 Gather Session 2 — 5 Nov @ 18:00 🔗 Full text + Code + Demo: tinyurl.com/kraxhpjd

2025.emnlp.org

November 4 –9Suzhou, China

111

Thomas Hueber @thueber.bsky.social · 22/09/2025

Had the pleasure of visiting Okko Räsänen’s lab at Tampere Univ. Gave a talk on our work on computational models of speech acquisition and served as the opponent in the PhD defense of María Andrea Cruz Blandón, an impressive body of work. Congrats to her & the group, a great place for research!

021

Thomas Hueber @thueber.bsky.social · 01/09/2025

🚨 Open PhD Position – Grenoble, France 🚨 Join us at GIPSA-lab to explore how Speech Language Models can learn like children: through physical and social interaction. Think AI, robots, development 🧠🤖🎙️ Fully funded (3 yrs) • @cnrs.fr / @ugrenoblealpes.bsky.social Details 👉 tinyurl.com/bde988b3

tinyurl.com

087

DevAI&Speech involves researcher and engineers from GIPSA-lab (CNRS, Université Grenoble Alpes), Laboratoire de Psychologie et de NeuroCognition (Mathilde Fort), Tampere University, Inria, and Atos (6/6)

010

📢 Several fully funded PhD positions will be announced soon — but feel free to reach out already if you’re interested! (5/6)

🤖 embedding SpeechLMs in our humanoid robots and training them through natural interaction with humans 👶 better understand the underlying mechanisms of speech acquisition through experimental studies involving parents, children, and robots at Grenoble’s Babylab. (4/6)

Key goals include: 🧠 integrating knowledge of biomechanics in SpeechLMs 👁️ enabling SpeechLMs with multimodal input/output processing (3/6)

We’ll be developing Speech Language Models (SpeechLMs) that learn like children do: through multimodal sensory input (audio, images) and interactive experiences with both their speech production system and social environment. (2/6)