Qwen3-ASR Open Source: Stable, Streaming, and Multi-language Support!

Alibaba Cloud's Qianwen team has open-sourced the Qwen3-ASR series of speech recognition models, including 1.7B/0.6B parameter versions and a forced alignment model. This series supports 52 languages and dialects, covering 22 regional Chinese accents and multiple English variants, and performs stably in noisy environments and singing scenarios.