(Yicai Global) Dec. 11 -- Chinese artificial intelligence start-up Yitu Technology has reached a technical cooperation agreement with Microsoft and Huawei Technologies, Yitu Technology said in a statement today.
The Shanghai-based artificial intelligence technology innovator will provide voice technology to third-party application developers and release speech-to-text application programming interfaces for short audios, the company said.
Yitu and Microsoft will cohost a voice technology open platform based on Redmond, Washington, US-based Microsoft's Azure Cloud and grant a wide range of third-party app developers access to its speech recognition technology, per the official press release.
Yitu and Shenzhen-headquartered Huawei will release the Intelligent Voice Joint Solution, which uses Yitu's open platform of voice technology and Huawei's Ascend series chips and Atlas 300AI accelerator cards to form an integrated solution for hardware and software.
Yitu will release long audio speech-to-text APIs and real-time speech-to-text APIs in future.
The firm has amassed reams of real dialog data and specific corpora of professional and life classes to improve its speech recognition accuracy.
The company came up with its own multi-dimensional and multi-scene test data set to train and trial its model algorithm, Lv Hao, Yitu's chief innovation officer and a former Google Research Scientist, told online tech media 36Kr.
The incorrect word-recognition rate of Yitu's speech recognition for short audios is 3.71 percent in the AISHELL-2 test, which is based on the world's largest open source Chinese database, 20 percent less than that of the original sector leader iFlytek, Yitu Technology said.
Formed in 2012, the firm focuses on facial and speech recognition. The company has achieved progress in commercializing its security, healthcare, finance and retail products.
It will now focus on accumulating technology, Lv said, adding that it provides its own API and model algorithms on an open platform to test them in the market and thus reach different sectors and scenarios.
Editor: Ben Armour