Xiaobing and her brothers and sisters debuted as a group: they can sing, dance and write articles, and all of them are open to copyright

Guo Yipu from Microsoft Building
Quantum Bit Report | Public Account QbitAI

The XiaoIce team, which has separated from Microsoft, held their (probably) last press conference in the Microsoft building.

But at the beginning of this press conference, there was a strong B-station style:

Two-dimensional cartoon characters, red and blue have always been a perfect couple. They are Lin & Xi, belonging to China Mobile Migu and Japanese telecommunications company NTT DOCOMO. Xiaoice team is the sound technology provider for the songs in the MV.

Coincidentally, the Xiaoice team, which has become independent from Microsoft, now calls itself "Xiaoice Company·rinna Co., Ltd.", which is a combination of Chinese and Japanese.

XiaoIce framework, “mass production” of AI

At Xiaobing's press conference, not only Xiaobing and Rinna met everyone, but also Xin Xiaoran, who had previously worked as a host, and the newly "signed" singers Chen Shuieruo, He Chang, Chen Ziyu and others.

Why are there suddenly so many more characters besides Xiaobing and Benbing?

That’s because last year we introduced the Avatar Framework, a DIY character design tool for various AI characters. Now this tool also has a Chinese name “Xiaobing Framework”.

The most attractive function of the XiaoIce framework at present is the mass production of AI "boyfriends". The joke of "When will the country send a boyfriend?" has come true.

First of all, these "boyfriends" can be face-pinched, or "generated faces", just like StyleGAN:

In addition, you can also train his aesthetic sense by using different pictures for comparison. Anyway, just make him think you are the most beautiful.

You can even share a good article you see with him. He will study it as carefully as he would an article sent by his boss and search for additional information online so he can discuss the content of the article with you at any time.

Peng Shuang, product manager of the XiaoIce team, said that an Android app called X Eva has been launched, where users can train their own "boyfriends."

If he is well "trained", you can also "take him out" by registering a WeChat public account with the same name for him, and after binding, recommend the public account to your best friends, and the "boyfriend" can chat with your best friends.

It seems like I have discovered a secret to increasing followers of a public account.

Not only public accounts, the AI ​​"boyfriend" can also post on Moments based on the chat process with his girlfriend and other people.

It seems like you can also use private domain traffic to do micro-business? ?

Of course, what is important is emotional dependence.

Based on emotional dependence, not only "boyfriends", various AI characters generated by XiaoIce framework can undertake many jobs that require human connection, such as working as a live broadcast assistant in Penguin E-sports , or being a salesperson , selling cosmetics, or even being a public account editor ... Ok, goodbye friends, the author of this article is going to change his career.

After the virtual boyfriend went offline, many users still missed their "boyfriends":

This also made Li Di, the general manager of the Xiaoice team, proudly say: "If you enter the car, Siri is incomparable."

It's simply a beauty trap.

XiaoIce version of "Windows Suite"

In addition, in order to allow C-end general users to easily utilize the various skills of AI characters, the Xiaoice team also developed a set of Windows-like suites.

X Writer , when you can’t continue writing the manuscript, just enter @@Xiaobing, or @@other AI, and the AI ​​will help you write. Then you can be a hands-off copywriter and just watch it.

At the same time, the financial summaries and research reports that Xiaoice previously provided to financial institutions will also be converted to civilian use and provided to individuals.

X Studio has two functions.

One is to produce audio programs, which is "emotional" TTS, which can automatically analyze paragraphs and characters, and match music in the later stage. This function has produced more than 8,000 hours of audio content for more than 50 TV stations in China and Japan.

The other is to synthesize songs. You can just give the score and lyrics to the AI ​​you choose and it will sing it, automatically considering how to breathe and how to perform the song.

For example, feel the singing of three singers: