Making your own virtual digital people with AI at low cost? to also go

Although the concept of virtual digital human has been widely discussed, there is no unified and complete definition, but it is generally acknowledged that digital virtual human is driven by artificial intelligence and other technologies. For example, we can drive a character or image in an image or make it into a new video. So, based on this basic cognition, can we be a simple and low-cost virtual digital person? Sure, that's good.

The concept of low cost is to prevent professional engines such as virtual engines from recreating independent and complete personal images, and only use basic artificial intelligence technology to drive the characters in an image to speak. In this way, it is much simpler. And this technology, not to mention now, even a while ago, has come out, just integration.

Simply put, it is to take a photo or make an image with yourself (the character), and then drive it with AI to let the character speak. This AI is often not technology. If the figures in the picture want to speak, they have to speak first. You can do this with an AI program like wav2lip. Next, since you said it, what about this "speech", that is, how to interpret the voice content, you can record it yourself-haha, of course, the witty guest is joking. If you shoot it yourself, it is not a virtual digital person, but a real person. Of course, the content of this language needs to be integrated. And this requires another AI program, speech synthesis.

This technology is open source by Baidu Flying Paddle. PaddlePaddle's own voice project, which is open source, can be described as a heavy blow. Very heavy, it allows our individual developers to call their trained models to do their own speech synthesis and some other projects. Wit I have said in the previous article, so I won't say much. Speaking of this speech synthesis application, there is only one female voice here for the time being. You can't switch other voices after using it.

In fact, as for low-cost virtual digital people, friends who may be concerned about artificial intelligence also know that Baidu has launched an open source project, which is to use such technology to realize the simple production of a virtual digital person. In fact, this realization process is similar to that described by the witty guest above. I made an artificial intelligence version of Andy Lau blind date video before, which is similar to such a low-cost virtual digital person. On the other hand, the key point of this open source project on PaddlePaddle is that the open source author integrated various independent project applications of Baidu PaddlePaddle and made such a virtual digital person.

How about seeing teams or big factories launch their own virtual people this year (Lunar New Year)? I feel a little itchy. Do you want to be a virtual person of your own? If the budget is limited and the cost is limited, you can use this idea and the corresponding AI technology to make your own virtual human.