yxdu/mcat-large

MCAT: Scaling Many-to-Many Speech-to-Text Translation with MLLMs to 70 Languages

This repository contains the MCAT (Multilingual Cost-effective Accelerated Speech-to-Text Translator) framework, presented in the paper MCAT: Scaling Many-to-Many Speech-to-Text Translation with MLLMs to 70 Languages.

MCAT addresses key challenges in Speech-to-Text Translation (S2TT) tasks with Multimodal Large Language Models (MLLMs), specifically language coverage and efficiency. It introduces a language scaling method leveraging curriculum learning and a data balancing strategy to extend translation capabilities to 70 languages. Furthermore, an optimized speech adapter module is designed to reduce the length of speech sequences, enhancing batch inference efficiency.

For more details, including training and evaluation scripts, please visit the GitHub repository.

✅ Current Version MCAT (v2.0)

• Supported 70 Languages: Afrikaans (afr), Amharic (amh), Arabic (ara), Assamese (asm), Azerbaijani (azj), Belarusian (bel), Bengali (ben), Bosnian (bos), Bulgarian (bul), Catalan (cat), Czech (ces), Chinese (cmn), Welsh (cym), Danish (dan), German (deu), Greek (ell), English (eng), Estonian (est), Persian (fas), Finnish (fin), French (fra), Galician (glg), Gujarati (guj), Hebrew (heb), Hindi (hin), Croatian (hrv), Hungarian (hun), Armenian (hye), Indonesian (ind), Icelandic (isl), Italian (ita), Javanese (jav), Japanese (jpn), Kannada (kan), Georgian (kat), Kazakh (kaz), Khmer (khm), Kyrgyz (kir), Korean (kor), Lao (lao), Latvian (lav), Lithuanian (lit), Malayalam (mal), Macedonian (mkd), Malay (msa), Burmese (mya), Dutch (nld), Norwegian (nob), Nepali (npi), Punjabi (pan), Polish (pol), Portuguese (por), Romanian (ron), Russian (rus), Slovak (slk), Slovenian (slv), Spanish (spa), Serbian (srp), Swedish (swe), Swahili (swh), Tamil (tam), Telugu (tel), Tagalog (tgl), Thai (tha), Turkish (tur), Ukrainian (ukr), Urdu (urd), Uzbek (uzb), Vietnamese (vie), Cantonese (yue)
• 4830 Translation Directions - Supports all 4830 possible translation directions (70×69 language pairs)

Installation

conda create -n m2m-70 python=3.10
conda activate m2m-70

git clone https://github.com/yxduir/m2m-70
cd m2m-70/SLAM-LLM

sudo apt update
sudo apt install ffmpeg
sudo apt install git-lfs

pip install -r requirements.txt
pip install -e .
cd ..

Download Model

Encoder	Adapter	LLM
Whisper-large-v3	Adapter	Gemma-3-27b-it

cd models/
# Total 150G of storage space for models
git lfs clone https://huggingface.co/openai/whisper-large-v3
git lfs clone https://huggingface.co/yxdu/mcat-large
# Access to the Gemma models is required before using git lfs.
git lfs clone https://huggingface.co/google/gemma-3-27b-it
cd ..

Infer Demo

This is a demo inference script, covering translation between 70 languages, with a total of 70×69=4,830 directions. This demo downloads the 9 GB dataset from HuggingFace. It requires GPUs with 80GB VRAM, with support for BF16 only.

bash scripts/infer_demo.sh

Citation

@misc{du2025mcatscalingmanytomanyspeechtotext,
      title={MCAT: Scaling Many-to-Many Speech-to-Text Translation with MLLMs to 70 Languages}, 
      author={Yexing Du and Kaiyuan Liu and Youcheng Pan and Bo Yang and Keqi Deng and Xie Chen and Yang Xiang and Ming Liu and Bin Qin and YaoWei Wang},
      year={2025},
      eprint={2512.01512},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2512.01512}, 
}
@article{du2025speech2text,  
  title     = {Making LLMs Better Many-to-Many Speech-to-Text Translators with Curriculum Learning},
  author    = {Du, Yexing and Pan, Youcheng and Ma, Ziyang and Yang, Bo and Yang, Yifang and Deng, Keqi and Chen, Xie and Xiang, Yang and Liu, Ming and Qin, Bing},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)},
  year      = {2025},
}