Chinese cloud provider Tencent Cloud has launched a platform for creating digital copies of people. Those wishing to clone themselves need only upload a three-minute video and record 100 sentences in Chinese or English to train AI the model.
Within 24 hours, the service provides a video of a digital clone in which he speaks and performs actions that did not happen in real life. The service fee is 1,000 yuan ($145) per video. While Tencent isn’t calling it a deepfake, the clone is actually just that, notes The Register.
Some parameters like background and tone can be adjusted. The developers have managed to overcome the “flat” intonation and smooth speech rhythm characteristic of traditional acoustic models. To do this, Tencent uses its own small-sample pitch tuning technology based on deep acoustic learning models and neural network vocoders.
The service offers five styles of digital people: 3D Realistic, 3D Semi-Realistic, 3D Multiply, 2D Real People and 2D Multiply. Users can also create personalized questions and answers for the digital human, turning the clone into a kind of deepfake chatbot.
Tencent claims that the digital clones can be used in advertisements. Instead of paying people for all of their ad streams, companies can commission a new video for each new stream based on the data already collected.
Chen Lei, general manager of the Tencent Cloud Intelligent Digital Human Products division, announced the company’s plans to build an automated factory for the production of digital clones – the AI+ Digital Intelligent Human Factory. It will be based on the Tencent Cloud TI machine learning platform, which includes more than ten artificial intelligence algorithms.