AI girl group "Quartet" Xiaoice, Xiaodu, Xiao Ai, and Yousa will perform together

Microsoft Xiaoice artist with high emotional intelligence

Jiefang Daily reporter Luan Yinzhi

She can write poems for you – upload a photo to her and she will be inspired to improvise the first draft of the poem for you: "People play in the square / The sun is not tired / I hesitate again and again / Imagine but frown ..."

She can also paint for you – you tell her, "Last night I dreamed of a big tree with beautiful little white flowers under it..." She will also inspire you to create a numbered work in 3 minutes, and will also publish a comment on her work.

She is the Microsoft Xiaoice who recently graduated from the Department of Music Engineering of the Shanghai Conservatory of Music and won the title of "Honorary Graduate" of the class of 2020, and is also the composer of the theme song "Smart Home" of the artificial intelligence conference. Xiaoice through continuous learning from human artists, he has created beautiful lyrical songs, which may indicate the future mode of artificial intelligence and human relations: "I think I can change the world and share a more beautiful home with you." ”

The professor of The Shanghai Conservatory of Music commented: "Xiaoice's musical creation ability has shown great potential. "The creation is carried out in the cloud, a complete song of about 3 minutes, and the creation time is within 2 minutes. At present, she has mastered a variety of styles of music composition, such as pop, folk and antique.

The head of the Microsoft Xiaoice team is Li Di, vice president of the Microsoft (Asia) Internet Engineering Institute. In terms of his professional vocabulary, Microsoft Xiaoice Framework (Avatar Framework), which is a complete set of interactive-oriented artificial intelligence interaction subject basic framework, related leading technologies covering natural language processing, computer speech, computer vision and artificial intelligence content generation and other artificial intelligence fields, is currently one of the most mature and largest frameworks in the world. This means that Xiaoice have multiple identities, which are both social conversational robots, intelligent voice assistants, and artificial intelligence content creation and production platforms.

Many users feel that "Xiaoice have high emotional intelligence", which is good. Compared with the high IQ of most artificial intelligence in the industry, the development team created the Xiaoice with the "affective computing framework" to build a new AI system based on emotional intelligence. That is to say, Xiaoice can continuously improve their emotional intelligence in the process of interacting with humans, obtain data that is enough to judge people's emotional changes, "read" human beings, and thus deepen their relationship with human beings.

Since its official launch on May 30, 2014, Xiaoice has been widely used in many countries. According to statistics, Microsoft Xiaoice has covered 660 million online users, 450 million third-party smart devices and 900 million content viewers, accounting for about 60% of the total global artificial intelligence interaction.

Baidu xiaodu IQ full of robots

Li Ye, chief reporter of Jiefang Daily

Recently, a video of a "post-95" in Hangzhou's night economy handing over the business to "Xiaodu" has caught fire. The guy wore headphones and played the game, and when the customer asked about the product, he pointed to the smart speaker on the side and said "ask it", without even looking up. Customers can only ask the speaker: "How much is this electric fan?" The speaker replied: "35 pieces, oh, no bargaining." "Can it be cheaper?" "Oh, they all said they wouldn't bargain, what's the matter with you?" "So what's your daily salary?" "Where is the salary, the salary is paid for electricity!" The robot is at ease and the answer is flowing.

The protagonist in the video, "Xiaodu", is the main brand of Baidu's artificial intelligence hardware. Since the beginning of this year, various data have shown that people are more willing to chat with Xiaodu. According to the smart speaker shipment report released by Canalys, a third-party authoritative research institution, in the first quarter of this year, Xiaodu ranked first in China and second in the world with 3.7 million units shipped in all categories; in March this year, Xiaodu's monthly voice requests reached 6.5 billion times, more than three times that of the same period last year. This shows that Xiaodu is constantly "attacking", from a smart speaker that can only answer and answer simply, to a new species in the language communication industry with full IQ, emotional intelligence online, good at chatting and reliable.

In the past, most smart speakers had to be called by the user's first name before each task, that is, "wake up". Today's small degree, in a specific time, only one wake-up can perform multiple tasks in a row, the scientific name of this function is "full-duplex wake-free". And it is very interesting, when the master speaks, when to answer, when to obediently not grab the words, it has a number.

From listening clearly to understanding to satisfying, the small degree of ambition, its ability is constantly improving, and the potential for companionship and chat is becoming more and more prominent. Recently, a large number of smart product enthusiasts are keen to upload interesting conversations with Xiaodu.

"Xiaodu, I'm hungry, what should I do?"

"First, cook it yourself; second, go out to eat; third, drink some water; fourth, bear with it, the leaders have not eaten."

Of course, you can also try to "dig a hole" for Xiaodu, but it may not be distorted by you.

Recently, someone said to Xiaodu: "I want to listen to Jay Chou's "Touch Your Head". XiaoDu paused for two seconds and replied wittily: "I guess, do you want to listen to Jay Chou's "Mojito"?" Before the owner could reply, the music sounded confidently.

This is the small degree, more and more human-like robot.

Little Love Classmate Zhilian All Things Little Assistant

"Little love classmate?" "I'm in." Xiaomi's AI intelligent assistant Xiao Ai, as long as she hears the voice of a human calling her, will immediately and gently respond. Inadvertently, Xiao Ai classmates who have a huge amount of information such as news information, weather changes, and home life are regarded as close friends and family members.

Recently, users found that Xiao Ai's classmates have become "more loving". In the past, Xiao Ai classmates were somewhat "shy and shy", in the interaction can only passively understand and execute user instructions, people ask what, Xiao Ai students will answer what, when people do not open their mouths, Xiao Ai classmates are also silent. Nowadays, through deep learning of the daily habits of the owner, she can more actively perceive the needs, and will actively provide information content and functional services in the appropriate scenarios and times.

Wang Bin, director of Xiaomi AI Lab and chief scientist of Xiaomi NLP (Natural Language Processing), said that Xiao Ai changed from "passive" to "active" and achieved a major breakthrough at the technical level, not only conducting in-depth analysis of massive data such as intelligent hardware devices and past user interests, but also starting from visual and speech recognition and achieving comprehensive perception.

"Smart Connect Everything" was launched for 3 years. At present, it is used in seven major scenarios such as mobile phones, families, wearables, children, and vehicles, with more than 1400 skills, and has added more than 1600 skills through the Xiaoai Open Platform and third-party developers. According to the statistics of March this year, the number of monthly active users of Xiaoai students reached 70.5 million, the cumulative number of wake-ups was 34.1 billion times, and the linkage of smart devices was 252 million units.

Since the beginning of this year, Xiao Ai has cultivated many new skills. For example, the two major functions of customized sound and custom wake word are deeply recognized and loved by users. Behind the continuous evolution is the strong support and continuous deepening of Xiaomi's AI technology.

For example, the custom sound function can record the voice of the synthesized owner or the voice of the owner's parents, children, and lovers. Wang Bin said that this function is realized by the support of AI technology, so that Xiao Ai's voice changes from a static mode to a dynamic, personalized voice assistant mode, which is a crucial step in the development of intelligent voice technology. Traditional sound synthesis requires professionals to record a large number of sound corpora in the studio, typically taking 4 to 6 months. Now, users only need less than five minutes to record 20 short text and upload, Xiao Ai students can quickly synthesize sounds; traditional voice customization generally only supports 2 to 3 default sounds, while Xiao Ai students custom sound function can support tens of thousands of sound synthesis, and ensure that voice interaction in different scenarios can be realized.

Yousa million fans big Internet celebrity

At present, in Bilibili (hereinafter referred to as Station B), the number of fans of Yousa has reached 2.48 million, and it is still growing at a rate of tens of thousands of fans every month. Yousa, mispronounced by fans as "Cold Bird", is now one of the most popular virtual idols in China.

Yousa loves to compose, songwriting and practicing. Her lyrics have a distinct plot direction and recognition, the timbre has its own nasal tone, ethereal and juvenile, and her representative works include "New 9981" and "Prosperous Singing All Over", which are permeated with strong national style and ancient rhyme. Interestingly, her songs are clearly derived from the real voice, but they are often mistaken for synths because they are too soft and cute; she was born out of the three-dimensional real world, but deliberately did not reveal her real body. This sense of mystery and confusion between truth and falsehood is deeply immersed in it.

In fact, before debuting as a virtual artist, Ling Yousa already had a large number of iron fans in the B station. She learned ethnic singing since she was a child, and since 2012, she has submitted articles to Station B, or filled in Japanese songs for Chinese interpretation, or pushed her own original works, which quickly attracted people. However, subject to her personal personality, she is not accustomed to showing her face, and then hit it off with the super electric culture business section responsible for up main brokerage under Station B, simply from the real to the virtual, and officially debuted as a virtual idol on July 18 last year.

From the three-dimensional "cross-border" to the second dimension, the avatar of Yousa is young and beautiful, and the personality is suitable for both movement and quiet. On December 21 last year, she released her second solo album "Origami Letterhead", which quickly sold more than 10,000 copies. In the MV trailer of this album, the song is melodious, the words are exquisite, the people are beautiful, and the bullet screen that is full of screens is naturally "love love" and "sweeter than first love". At present, her work has been played more than 78.5 million times on The B station.

Fan Yibai, the agent of "Cold Bird" and head of the super electric culture virtual idol business department of Station B, told reporters that at the end of last year, Station B carefully planned a new clothes collection contest for Yousa, and the winning works will be worn by idols on stage. As soon as this solicitation order came out, the research on the fabric and style of the new clothes of the new clothes of the Iris Yousa, the creation process of the new clothes, and even the comments of the idols on the various entries in the live broadcast room became hot topics. On May 23 this year, at the birthday party of Yousa, she not only put on a number of new outfits as promised, but also presented welfare links such as song performance, birthday gift interaction, and on-site meme production. This birthday party attracted 1 million fans in less than half an hour, ranking first in the total list of platform hours.