Repository navigation
support qwen3-asr in speech to text #769
Description
Problem description
Qwen has recently open-sourced a new ASR model, and its performance is somewhat better than Whisper. hope it can be supported. this is repo: https://github.com/QwenLM/Qwen3-ASR?tab=readme-ov-file#python-package-usage
Proposed solution
https://github.com/QwenLM/Qwen3-ASR?tab=readme-ov-file#python-package-usage
Alternative solutions
No response
Benefits to React Native ExecuTorch
This would significantly improve the accuracy and stability of speech transcription in React Native ExecuTorch, while enabling robust multilingual support.
Additional context
https://github.com/QwenLM/Qwen3-ASR?tab=readme-ov-file#python-package-usage
Activity
mkopcins commentedon Jan 30, 2026
HI @ryzencool! Yes, we are saw the model and we are quite excited about this option, but the problem might be the performance, although it is based on small 0.6B variant, which performs very well as LLM, both S2T as well T2S add significant compute requirements.
TLDR we have to test perfomance on device
ryzencool commentedon Feb 2, 2026
@mkopcins Thank you for your response — I’m really excited and looking forward to it.