Português
EnglishDingTalk and Tongyi Lab Launch Industry-Specific Speech Recognition Large Model Fun-ASR
On August 22, DingTalk and the Tongyi Lab speech team jointly released a new generation of speech recognition large model, Fun-ASR. The model can accurately recognize professional terms in ten industries, including construction and animal husbandry, and supports custom training of enterprise-specific models. Leveraging deep collaboration, Fun-ASR efficiently transcribes various audio signals and features multi-industry term understanding, multilingual accent recognition, and contextual semantic reasoning capabilities.
Currently, Fun-ASR has been integrated into DingTalk’s meeting captioning and simultaneous translation, smart minutes, voice assistant, and other functional modules. It is designed to build a stable, efficient, and scalable speech recognition foundation, particularly suitable for enterprise scenarios with high requirements for contextual understanding and recognition accuracy.
Core Technical Highlights: Three Key Capabilities Underpin High-Precision Recognition
Fun-ASR is trained on hundreds of millions of hours of audio data and co-built with real-world scenario data from DingTalk across multiple industries, including Internet, technology, construction, animal husbandry, and automotive, significantly enhancing its ability to recognize specialized terminology.
Test results show that recognition accuracy in the insurance industry has improved by 18%, while industries such as construction and animal husbandry have seen improvements of 15%-20%. The model also supports enterprise-defined hot words, allowing up to 1,000+ proprietary terms to be imported, enhancing recognition of rare or niche terms.
Fun-ASR can combine internal information within an enterprise’s DingTalk, such as the contact list, calendar, and knowledge base, to optimize reasoning. With enterprise authorization, it effectively reduces hallucinations in large models, delivering more reliable transcription results.
Based on an efficient end-to-end architecture, the model can further optimize its algorithms using actual voice data provided by enterprises, improving recognition accuracy for brand names, project codes, product names, and personal names, among other proprietary content.
Take Gujia Home Furnishing as an example: After dedicated training, the model can accurately recognize complex phrases such as "Belgian-imported Pulse latex" and "patented Sonocore foaming technology," providing a solid foundation for subsequent customer needs analysis.
Future Outlook: Continuously Deepening Industry Adaptability
Li Xiangan, head of the Tongyi Lab speech team, said: "We look forward to working with DingTalk to drive innovative applications of speech recognition technology in enterprise settings. In the future, we will continue to expand the data and model scale of Fun-ASR, enhance the replicability of our solutions, and deliver a smarter, more efficient experience for enterprises."
Zhu Hong, DingTalk's CTO, noted: "In just three months of close collaboration, we achieved model deployment and gained recognition from leading customers. This is a key breakthrough toward industry leadership and provides a replicable example for more enterprises to customize large models."
Currently, the potential of Fun-ASR continues to be explored. Both parties will focus on advancing upgrades in areas such as dialect recognition, robustness in noisy environments, multilingual support, and deep enterprise customization, comprehensively enhancing the precision and practicality of speech transcription to help more enterprises achieve intelligent upgrades.
DomTech is DingTalk's officially designated service provider in Macau, specializing in providing DingTalk services to a wide range of customers. If you'd like to learn more about DingTalk platform applications, you can directly consult our online customer service or contact us by phone at +852 95970612 or by email at cs@dingtalk-macau.com. We have an excellent development and operations team with extensive market service experience, ready to provide you with professional DingTalk solutions and services!