(Yicai Global) Oct. 25 -- Baidu Inc. [NASDAQ:BIDU] rolled out a new artificial intelligence (AI) system for DeepVoice, which can generate simulated human voices from text using deep learning technology. The company previously released two versions, and the final update is more efficient and powerful than its predecessors.
The new version of DeepVoice claims it can learn more than 10,000 voices in half an hour, which opens doors for many applications, cnBeta reported.
This technology can give different characters in video games and audio books their own voices to enhance the user experience. Compared to human voices, those made by DeepVoice sound artificial.
The first edition of DeepVoice could only process short sentences, and could learn one accent at a time after several hours. DeepVoice 2, which Baidu released in May, could imitate an accent within an hour and the system was able to learn hundreds of different accents.
Baidu is not the only search engine company working on text-to-speech, and Alphabet Inc.'s [NASDAQ:GOOG] DeepMind has been pursuing similar projects such as WaveNet. It boasts better accents and can produce human-like 'labial' sounds that have been used in the English and Japanese versions of Google Assistant.