Audar-ASR-V1:Arabic-first speech recognition foundation models with open weights

Hi everyone, We're releasing Audar-ASR-V1, the first public release from AudarAI.

AudarAI is building multilingual audio intelligence, beginning with speech recognition for Arabic and its many dialects.

We start with Arabic—not because it is our destination, but because it is where we believe we can make the greatest impact. From there, we continue expanding toward multilingual audio intelligence for the world.

This release includes: • Flash & Turbo open-weight models • Technical report • Evaluation harness • GitHub repository • Online demos

We'd really appreciate feedback from researchers and developers. If you find issues, have suggestions, or want to benchmark against your own systems, we'd love to hear from you.

Blog https://www.audarai.com/news/introducing-audar-asr-v1

GitHub https://github.com/AudarAI/Audar-ASR-V1

Hugging Face https://huggingface.co/audarai

2 points | by AudarAI 16 hours ago

2 comments

  • ankur64 5 hours ago
    This is cool. Nice wedge and I’d love to see more products like this. I have a friend who relies on Chinese models for mandarin based apps because they handle special characters better than classic foundation models (understandably). Best of luck with the launch!
  • Sia123 16 hours ago
    It sounds really great! Prioritizing Arabic from the very start is a major breakthrough for multilingual audio intelligence in the region.