Announcing the MLC-SLM Challenge for AI Development
Nexdata, a prominent global provider of AI data services, is excited to unveil The Multilingual Conversational Speech Language Model Challenge (MLC-SLM). This challenge is officially recognized as a satellite event of the upcoming Interspeech conference, set to foster advancements in multilingual speech technology.
Objectives of the Challenge
This challenge is supported by major tech players such as Meta, Google, Samsung, Naver, and China Mobile. The key goal is to push the boundaries of multilingual conversational speech AI, offering a real-world dataset while promoting innovative approaches to developing speech language models.
Challenge Breakdown: Tasks Overview
Participants will engage in two distinct tasks focused on enhancing speech language models (SLMs):
Task I: Multilingual Conversational Speech Recognition
The primary objective of this task is to create a multilingual LLM-based automatic speech recognition (ASR) model. Participants will benefit from oracle segmentation and speaker labels to facilitate their development process.
Task II: Multilingual Conversational Speech Diarization and Recognition
This task requires the development of a complete system for speaker diarization, where the identification of who speaks when is crucial, alongside speech recognition that converts voices to text. Unlike the first task, participants will not receive any prior or oracle information during evaluation, encouraging a more flexible approach using either pipeline-based or end-to-end systems.
Rich Language Data Resources
The training dataset includes around 11 diverse languages: English, French, German, Italian, Portuguese, Spanish, Japanese, Korean, Russian, Thai, and Vietnamese. This comprehensive collection aims to offer rich resources for training and evaluating multilingual conversational speech language models (MLC-SLM), addressing significant challenges like linguistic diversity and speaker variability.
Key Dates for Participation
Participating teams are encouraged to note the important timeline:
- Registration opens: March 10, 2025
- Training data release: March 15, 2025
- Development set and baseline system release: March 20, 2025
- Evaluation set release and Leaderboard open: May 15, 2025
- Leaderboard freeze and paper submission portal opens: May 30, 2025
- Paper submission deadline: June 15, 2025
- Notification of acceptance: July 1, 2025
- Workshop date: August 18, 2025
Prize Distribution for Top Performers
To encourage participation, Nexdata has allocated a total prize pool of $20,000 to be awarded to top teams in each task, with the distribution as follows:
- 1st Place: $5,000
- 2nd Place: $3,000
- 3rd Place: $2,000
Get Involved!
To find more information about the MLC-SLM Challenge, please visit the official challenge website. Participants can register and get involved, contributing to the exciting developments in AI.
Contact Us
If you have any questions or require additional assistance, feel free to reach out. We're here to help you on this journey towards enhancing multilingual conversational AI.
About Nexdata
Nexdata specializes in providing exceptional training data solutions and is committed to supporting innovation in AI development. With a wide range of ready-to-use datasets and flexible data collection and annotation services, Nexdata aims to unleash the true potential of AI and facilitate growth in the AI sector.
Frequently Asked Questions
What is the MLC-SLM Challenge?
The MLC-SLM Challenge is an event to advance multilingual conversational speech AI, inviting participants to develop innovative speech language models.
Who are the sponsors of the challenge?
The challenge is supported by leading tech companies, including Meta, Google, Samsung, and others.
What are the major tasks participants will perform?
Participants will focus on two tasks: building a multilingual ASR model and developing systems for speaker diarization and recognition.
What is the prize for the winners?
The total prize pool for the challenge is $20,000, with cash prizes for the top-performing teams.
How can I participate in the challenge?
Interested individuals can find participation details on the challenge's official website and register to contribute.
