"Hello old and new friends of JD.com, I'm Liu Qiangdong. On the evening of April 15, Jingdong officially released a video in which Liu Qiangdong spoke in front of the camera. And Brother Dong in front of the camera looks so radiant, I believe that if it were not a public report, everyone would never have imagined that it was not Brother Dong himself behind the camera, but a high-fidelity "digital person"!
And the video released on the 15th is actually a "warm-up" for the official live broadcast: at 6:18 pm on April 16, at this specific point in time, the "Procurement and Sales Dongge" AI digital human created by Jingdong Yunyanxi opened the live broadcast debut, and at the same time appeared in the live broadcast room of Jingdong Home Appliances and Jingdong Supermarket.
Strictly speaking, this is not the first time that Dongge has appeared in front of the camera: according to China News Network, seven and a half years ago, on November 10, 2016, on the eve of "Double Eleven", Jingdong began to join forces with Sichuan Pepper at 8 o'clock in the morning to launch a special 12-hour uninterrupted live broadcast SHOW, and the domineering president Liu Qiangdong personally stood on the platform to cook live, and taught netizens to eat "big plate chicken" and "Boston lobster". At the same time, in the live broadcast, he did not forget to focus on recommending JD's products, emphasizing to everyone that "all ingredients and raw materials can be purchased from JD.com".
In fact, until the end of the official live broadcast, many netizens still couldn't believe that Dongge, who appeared on the scene, was really a "digital person", just because it was too realistic!
"Digital Man" Dongge, straight to the point?
To what extent is it realistic? According to the author's joint observation with friends, the "digital human" Dongge is not only speaking there, but also has rich body language and expressions, and at the same time, the tone of speech and intonation is also eight or nine percent of the restoration degree of Dongge's label-like "Suqian Mandarin"!
According to a number of media reports, Liu Qiangdong in real life speaks relatively fast, speaks softly, and some words will be used to reading continuously, and he pronounces the "sh" in "time" and "exactly" with a heavy nasal sound, and also likes to call "brothers" to boost everyone's morale, and these factors have been optimized in the "digital person" Dongge.
Of course, there is no doubt that there is a difference between AI and real people: according to the author's friend, the hand of "Brother Dong" in the picture is constantly moving, which is unnatural, and if you listen carefully to the tone of the voice, you can still hear the obvious synthetic breath, and the difference between it and natural speech is like the difference between the "semi-solid-state battery and the all-solid-state battery" that has been boiling recently.
In fact, the tone of the speech is not important, it is no problem if it can be understood, everyone is more concerned about what a big man like Liu Qiangdong will talk about during the live broadcast, is it his own entrepreneurial experience or chicken soup for the soul, or both?
Practice has proved that although "Brother Dong" did not become a chef again this time, nor did he sell ingredients, he still did his old job - bringing goods!
According to the live broadcast record, "I founded JD.com to make the goods you buy convenient, fast and cheap, and to ensure the quality of the products." After a simple 5-minute warm-up, the Dongge digital person quickly started bringing goods, and did not talk too much about his personal life and opinions. The products with goods are mainly home appliances and food, including air conditioners, TVs, blueberries, milk, corn, etc., with the 4month16day"Jingdong Home Appliances Home TV Air Conditioning Super Category Day。
According to the summary of brokerage China, Jingdong Supermarket disclosed that the digital version of "Dongge" was broadcast for 30 minutes, and the number of viewers in the live broadcast room exceeded 10 million, and the number of views exceeded 20 million in the past 1 hour, and the average stay time of users during the live broadcast period reached 5.6 times the daily average. Within 40 minutes, the overall order volume of the live broadcast room exceeded 100,000.
During the live broadcast, the overall order volume increased by 7.6 times compared with last Sunday, and the turnover of Jingdong Supermarket's "10 billion agricultural subsidies" increased by 5.7 times compared with last Sunday. It can be regarded as a "good start" record!
Liu Qiangdong's participation in the live broadcast this time is mainly to further warm up JD.com's content ecology and short video creation. According to Tianyancha's intellectual property information, recently, Jingdong has applied for the registration of trademarks such as "Lao Liu Special Field", "Youjing Home Products" and "Round Head Price", and the international classification involves advertising sales, clothing, shoes and hats, etc., and the current trademark status is pending substantive examination.
The "digital human" live broadcast, with the large model as the backing
Behind the birth of digital human, it is inseparable from the deep technical cultivation and accumulation of JD Yunyanxi for many years. And "Yanxi" is the 100-billion-level model of JD Cloud, the real "brain" behind JD's "digital human"!
From the perspective of industrial application, the virtual anchor in the live broadcast room belongs to the service-oriented virtual digital human, which has a higher technical threshold than the identity-based virtual digital human, and needs to solve the problems of different scenarios in practical applications. From the perspective of the industry, digital humans have become the focus of the development of the live broadcast industry. According to Securities Daily, iiMedia Consulting data shows that in 2025, the market size and core market size of China's virtual human-driven industry will reach 640.27 billion yuan and 48.06 billion yuan respectively.
Compared with real people, the most obvious advantage of "digital human" is that it does not need to eat, sleep, or go to the bathroom, so it can theoretically be broadcast 24 hours a day without dead ends, taking care of everyone's time, and there is no need for venues, makeup, clothing and other expenses, which significantly reduces operating costs. What's more, digital humans can respond to user needs in real time and enhance user engagement. Therefore, it can complement the real streamer!
For a simple example, 6 a.m. to 8 a.m. is the rest period for live anchors, but it happens to be the "most painful" time for novice parents - after getting up to change the baby's diaper, they have to go out to work, so some maternal and infant brands will take advantage of this gap to let the digital human anchor carry out "moisturizing and silent" care!
But to make a "digital person" Dongge, the biggest difficulty is that Liu Qiangdong is a well-known public figure, and the public is extremely familiar with his image, voice, voice and other characteristics, so if the "production" deviates too much from the real person, it will inevitably lead to crazy diss from all walks of life, and then doubt the professional ability of the large model team behind it, whether it is a "silver-like pewter gun head", so the pressure of the team is not ordinary!
So, in order to create a flesh-and-blood "Dongge", what efforts has JD Yunyanxi made?
According to The Paper, 21st Century Business Herald, etc., in order to create a real "Dongge", the technical team made many adjustments to the large model: at first, the speech materials "fed" to the large model were too formal although they were passionate and explosive. To this end, they used the latest recorded chatter as the main material, including Liu Qiangdong's own travel experience, and then extracted the rhythmic characteristics of the 5-minute speech and poured it into the large model, and through continuous optimization, finally created a voice very close to himself.
After the timbre is reproduced, it is also necessary to capture the "paralanguage" of the sound, including the speed of speech, intonation, stress, gasp, etc. These paralanguages are originally sparsely distributed, and large models are not easy to capture the rules, but they are an important auxiliary force in judging the meaning of the language, and without paralanguage, the sound will lack emotion, appear too "correct" and cold.
The Yanxi team's approach is to disassemble the accents and intonation of voice samples into phonemes, use NLP (Natural Language Recognition) to make the model notice them more clearly, and use ASR (speech recognition) to capture intonation and tone changes, and comprehensively judge when to start speaking. Combining the above technologies, a digital human voice is generated that can converse fluently and freely.
It is understood that during training, the Yanxi voice model is "fed" with 50,000 hours of massive fresh voice data in order to intelligently match different live broadcast styles. As early as before "Brother Dong" was born, JD.com's digital anchors were already all over the platform, and their voices were enough to "confuse the real with fake", even similar to the voice of a cross talk actor!
According to public information, in this year's Spring Festival free time live broadcast, JD Yunyanxi Digital Man sold 40 million yuan of goods, and the average conversion rate of free time live broadcast increased by more than 30%. As of April 2024, more than 4,000 brands have used digital human live streaming on JD.com to replace real people to complete free time live broadcasts. They can increase the conversion rate of idle time by more than 30%, but the cost is less than 1/10 of the live broadcast.
The reason why Jingdong is heavy on digital live broadcast seems to have another consideration: compared with other e-commerce, it seems to be "congenitally insufficient" in live broadcasting, so it intends to enter the live broadcast e-commerce in a big way through new methods such as digital live broadcasting!
Jingdong e-commerce, catch up
On April 10, almost a week before the start of the broadcast of "Dongge", JD.com announced the "Double Billion" plan - it will invest one billion cash and one billion traffic to encourage more anchors and MCN institutions to settle on the platform, and at the same time, it was also reported that JD.com urgently promoted the recruitment of anchors in a way that lowered the threshold.
One billion is an astronomical amount in the eyes of ordinary people, but investing in the field of live broadcast may not be able to make much of a splash. Let's take a look at the comparison between JD and Ali: According to China Business News, on March 26 this year, Cheng Daofang, general manager of the content e-commerce division of Taotian Group, announced at the 2024 Taobao Content E-commerce Festival that "in 2024, Taobao Live will add 10 billion cash investment and 100 billion traffic, and real money will increase investment in content e-commerce", which is an order of magnitude higher than JD.com!
Indeed, as a late entrant to live broadcast e-commerce, JD.com lacks head anchors like Li Jiaqi and Wei Ya, so it is somewhat powerless to face the "winner-takes-all" situation. JD.com's 1 billion investment is mainly used to grab anchors and users. In order to compete for anchors, JD.com subsidizes talents in 20 fields such as digital 3C, home appliances and home furnishings, and mother and child, providing more exposure and traffic incentives for high-quality creators.
JD.com's move is both a subsidy and a digital person, can it help its live broadcast e-commerce to a new level? Let's wait and see.