IWSLT 2025 Model Compression Track Call for Participation
IWSLT 2025 Model Compression Track invites participation in optimizing large speech-to-text models for better deployment in resource-constrained environments, focusing on English-German and English-Chinese translations.
The International Conference on Spoken Language Translation (IWSLT) 2025 is inviting participation in the Model Compression Track. This track, held in conjunction with ACL 2025, focuses on optimizing large multilingual speech-to-text models for better accessibility and deployment in resource-constrained environments.
Participants are encouraged to apply various compression techniques to the Qwen2-Audio model (8.2B parameters, 16GB storage) for English-German and English-Chinese speech translation tasks. The evaluation will be based on model size and COMET translation scores, with human evaluation for top-performing models.
For more information, visit the IWSLT 2025 Model Compression Track webpage or contact iwslt_offline_task_submission@fbk.eu.
Important dates include:
- Test set release: April 1, 2025
- Submission of system outputs: April 15, 2025
- System description paper submission deadline: April 21, 2025
- Camera-ready deadline: June 1, 2025
- IWSLT conference in Vienna: July 31 – August 1, 2025
Tags: model compression, speech-to-text translation, IWSLT 2025, ACL 2025, Vienna, Qwen2-Audio, COMET scores