#United States #Monrovia #Nexdata #AI data services #speech technology

Nexdata Launches Multilingual Conversational Speech LLM Challenge

On March 19, 2025, in Monrovia, California, Nexdata, a renowned international leader in AI data services, officially announced the commencement of the Multilingual Conversational Speech Language Model (MLC-SLM) Challenge. This initiative is a notable satellite event for the forthcoming Interspeech 2025, aimed at propelling advancements in multilingual conversational speech AI.

The challenge is backed by prominent industry players, including Meta, Google, Samsung, Naver, China Mobile, and Northwestern Polytechnical University. The goal of this collaboration is to spur innovation in speech language models by offering a real-world dataset, thereby encouraging participants to develop cutting-edge speech recognition and understanding technologies.

Challenge Overview

The MLC-SLM Challenge comprises two key tasks that participants must accomplish:

Task I: Multilingual Conversational Speech Recognition

This first task focuses on developing a multilingual ASR (Automatic Speech Recognition) model using LLM (Language Learning Model) methods. Participants will be provided with specific oracle segmentation and speaker labels to facilitate their work.

Task II: Multilingual Conversational Speech Diarization and Recognition

In this second task, the aim is to construct a comprehensive system capable of speaker diarization (determining who is speaking when) alongside speech recognition (transcribing audio to text). Significantly, no prior information or oracle data will be available for evaluation, which presents a unique challenge as participants must design their systems with flexibility in mind, whether leveraging end-to-end frameworks or more traditional, pipeline-driven approaches.

A diverse training dataset has been created to encompass approximately 11 languages, including:

• - English (en)
• - French (fr)
• - German (de)
• - Italian (it)
• - Portuguese (pt)
• - Spanish (es)
• - Japanese (jp)
• - Korean (ko)
• - Russian (ru)
• - Thai (th)
• - Vietnamese (vi)

This vast array of languages is designed to tackle critical issues pertaining to linguistic diversity and speaker variability, while also enhancing contextual understanding across different languages and cultures.

Important Dates

The timeline of the challenge has been meticulously outlined as follows (in AOT time):

• - March 10, 2025: Registration opens
• - March 15, 2025: Release of training data
• - March 20, 2025: Development set and baseline system released
• - May 15, 2025: Evaluation set available, and leaderboard opens
• - May 30, 2025: Leaderboard freeze; paper submission portal opens via CMT
• - June 15, 2025: Paper submission deadline
• - July 1, 2025: Acceptance notifications sent
• - August 18, 2025: Workshop date

Nexdata has also established a prize pool of $20,000 for participants, with top performers in each category receiving monetary rewards:

• - 1st Prize: $5,000
• - 2nd Prize: $3,000
• - 3rd Prize: $2,000

Participate in the Challenge

For those interested in participating, further information is accessible through the challenge's official website at Nexdata. To register, potential participants can fill out the participation form available here. Should you have any inquiries, you can reach out to [email protected]

About Nexdata

Nexdata is a premier provider of high-quality training data solutions, committed to being a trustworthy partner in the AI space. With a broad selection of pre-existing datasets and adaptable data collection and annotation services, Nexdata is dedicated to unleashing the full potential of AI and propelling the industry's expansion forward.

Join us in revolutionizing multilingual conversational AI and be part of this exciting, pioneering challenge!