88Z AI, 6 July — Alibaba has launched the upgraded Fun-ASR-Realtime, a real-time speech recognition model with sub-second initial latency and accuracy comparable to offline systems. It supports 30 languages and 16 dialects, featuring context awareness and self-correction capabilities, outperforming competitors in dialect recognition tests.
At a glance
- Initial latency is in the hundred-millisecond range with accuracy near offline models, supporting 30 languages and 16 dialects.
- Features context understanding and self-correction, achieving an average dialect accuracy of 88.62%, leading competitors.
The update
Alibaba has officially released the upgraded Fun-ASR-Realtime, a large model for real-time speech recognition. The model achieves significant performance breakthroughs by keeping the initial character latency within the hundred-millisecond range while maintaining recognition accuracy comparable to offline models. In terms of coverage, Fun-ASR-Realtime supports recognition for 30 global languages and 16 Chinese dialects. Technically, the model incorporates advanced context understanding mechanisms, enabling it to self-correct based on linguistic context to improve overall recognition quality. Test data shows that its average character accuracy rate reaches 88.62% across 16 dialect recognition tasks. This performance leads over similar products from Volcano Engine and Tencent, demonstrating robust capabilities in handling multilingual and dialectal speech.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 千问大模型Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.