Qwen3-ASR
Alibaba's 1.7B speech recognition model — 52 languages, claims to beat Whisper-large-v3
Alibaba's 1.7B speech recognition model covering 52 languages, claiming to surpass Whisper-large-v3.
Voice AI: lifelike TTS, accurate ASR, and real-time voice agents.
Key features
- ASR and language identification for 52 languages and dialects
- Forced-alignment model aligning text-speech pairs in 11 languages
- Streaming and offline unified inference in a single model
- Supports long audio transcription and vLLM batch inference
Good for
- Developers adding multilingual speech recognition to apps
- Teams needing timestamp-aligned transcription pipelines
Stack & tags
PythonPyTorchvLLMTransformersGradioasrspeech-recognitionmultilingualforced-alignmentstreamingopen-weightstranscription