Challenge and Workshop on Multilingual Conversational Speech Language Model (MLC-SLM)

As AI continues to shape the way we interact with the digital world, the demand for inclusive and diverse language technologies is greater than ever. The Multilingual Speech Language Model Challenge (MLC-SLM), hosted by Nexdata and organized as part of the ICML 2025 Competition Track, addresses this need by encouraging the development of advanced automatic speech recognition (ASR) systems across a wide range of languages and dialects.

This global competition invites researchers, developers, and AI practitioners to build and fine-tune large-scale multilingual speech language models capable of accurately transcribing speech in multiple underrepresented languages. With more than 70 hours of transcribed data provided per language and over 300 hours of publicly available test sets, the challenge creates a unique opportunity to explore the frontiers of multilingual model training, transfer learning, and bias mitigation.

Participants will compete in two tracks: Full Dataset and Low-Resource, pushing the limits of performance in both high-data and constrained environments. This structure reflects real-world challenges where access to linguistic data can vary significantly between languages. The aim is to develop models that are not only accurate and robust but also scalable and equitable.

Beyond the competition itself, MLC-SLM is a call to action for the research community to build technologies that support linguistic diversity, digital inclusion, and cultural preservation. By participating in this challenge, teams will contribute to a broader mission of democratizing AI for all languages, paving the way for more inclusive global communication systems.

Explore more here.