DingTalk and the Tongyi Lab speech team announced the joint launch of a next-generation speech recognition model called Fun-ASR, capable of understanding industry jargon across ten major sectors including home decoration and livestock, while supporting enterprise-specific model customization training. Currently, Fun-ASR has been integrated into multiple DingTalk functional modules, including meeting subtitles and simultaneous interpretation, intelligent meeting minutes, and voice assistants.

From a technical perspective, the Fun-ASR speech recognition model features three core highlights: First, it significantly enhances recognition capabilities for industry-specific vocabulary. The model has been trained on over 100 million hours of audio data, combined with real-scenario collaboration from DingTalk's multi-industry clients, enabling accurate understanding of professional terminology across more than ten fields including internet, technology, home decoration, livestock, and automotive sectors.

Second, combined with DingTalk, it delivers stronger contextual awareness and understanding capabilities. Fun-ASR can leverage existing enterprise information within DingTalk such as contact lists, calendars, and knowledge bases for inference optimization, effectively mitigating hallucinations caused by large models and providing more reliable transcription results. This capability requires enterprise authorization to take effect.

Third, for enterprises with advanced requirements, it supports customized speech recognition model training. Based on an efficient end-to-end training architecture, the model can utilize real-scenario voice data provided by enterprises for further algorithmic optimization, improving recognition accuracy for proprietary vocabulary such as brand names, project codes, product names, and personal names.

Regarding this collaboration, Li Xiangang, head of the Tongyi Lab speech team, stated: "We are delighted to partner with DingTalk to jointly advance innovation and application of speech recognition technology in enterprise scenarios. In the future, we will continue expanding Fun-ASR's data and model scale, constantly improving the replicability of large model voice intelligence solutions to bring more efficient and intelligent product experiences to enterprise customers."

DingTalk CTO Zhu Hong also commented, "DingTalk and the Tongyi team achieved successful deployment of the Fun-ASR model in just three months of close collaboration, earning high recognition from leading clients. This represents a key breakthrough in our journey toward industry leadership and will provide a reference example for creating professionally customized large models for more DingTalk customers."