Alibaba Qwen upgrades real-time speech recognition model, supports 16 dialects and 30 languages

Alibaba Qwen has upgraded its real-time speech recognition model Fun-ASR-Realtime, which now supports 16 Chinese dialects and 30 languages. The API is available on Alibaba Cloud's Bailian platform. In tests covering eight major dialect regions, the model achieved an average character accuracy of 88.62%, leading over similar products from Volcano and Tencent in 12 dialect categories.

The model keeps initial word latency under 100 milliseconds, with streaming accuracy close to offline levels and seamless multilingual switching. For languages like Thai in East and Southeast Asia, recognition accuracy improved by 20%, making it suitable for overseas customer service and international conferences.

The offline version, Fun-ASR-Flash, has also launched. On AI evaluation platform Artificial Analysis, it achieved a word error rate of just 1.7%, ranking first globally.