A virtual character closely resembling a popular streamer's speech and mannerisms hosts a broadcast. Beyond reading prepared lines, it makes various facial expressions and gestures while engaging in emotionally rich communication with viewers.
Generative artificial intelligence (AI) is evolving beyond simple information delivery or content creation, such as producing text, images, and video, into a comprehensive medium capable of real-time interaction. Drawing on the vast data accumulated over time, "AI streamers" that share emotional connections and take over roles once reserved for humans have emerged as a new topic of discussion.
Live streaming platform SOOP (067160.KS) is pursuing service enhancements that integrate AI into live environments where streamers and users communicate in real time. Behind SOOP's confidence in offering AI-based interactive viewing services lies the abundant interaction data it has accumulated. Creating AI that converses as naturally as a human and understands context requires data containing the reactions and communications of actual users.
'Communication Data' Built on 115 Million Monthly Hours
While YouTube and over-the-top (OTT) platforms have grown around content consumption data such as viewing records, SOOP is finding unrivaled competitiveness in the real-time interaction data generated during live broadcasts.
About 15,000 streamers are currently active on SOOP, with around 7 million broadcasts conducted annually. According to global viewing data analytics firm Streams Charts, SOOP's total viewing time over the past 30 days was tallied at approximately 115 million hours.
Within these simultaneously running live broadcasts, various reactions accumulate in real time, including each streamer's voice, facial expressions, and gestures, as well as users' chats and donations. In fact, on SOOP, chats are exchanged at an average of about 400 million times per month and more than 150 times per second, with about 5 billion data points accumulating annually. This real-time behavioral pattern data holds value distinct from typical VOD consumption data and serves as a powerful driving force for SOOP to develop its own AI services.
In addition, SOOP is advancing its AI models based on roughly three years' worth of replay (VOD) and text data. Even now, it learns from about 200 million chat and post data points and about 1,000 hours of live and VOD data each day.
SOOP is using this data to develop its own "Large Video Model." This is a multimodal AI technology that comprehensively understands video, audio, and behavioral information, distinguished by its ability to learn a streamer's speech patterns, facial expressions, and even the mood of a broadcast. Through this, SOOP is attempting to implement AI that not only answers questions but fully reflects a particular streamer's individuality and identity.
'Customized AI Services' Filling the Streamer's Absence
Concrete experiments to apply customized AI to live environments are also active. Representative examples include the generative AI "SAVYG" for streamers, the video assistant "SOOPI" for users, and the AI manager "SARSA" supporting broadcast operations. These technologies support the live streaming ecosystem as a whole and ultimately focus on expanding interaction between streamers and users.
In particular, "SARSA 2.0," released on the 6th, learns a streamer's voice, facial expressions, and past broadcast data to assist in running broadcasts. Equipped with features that help content continue uninterrupted even when a streamer is away, it has drawn responses from users praising it as fresh.
"While most AI services focus on providing information or improving work efficiency, SOOP uses AI as a new communication tool that connects streamers and users," a SOOP official said.
In fact, these AI services have established themselves as a new culture among SOOP users. Users share AI-summarized broadcast content or conversations they have had with AI on the community, and enjoy the very experience of conversing with AI that reflects a streamer's personality as a form of play. It is currently estimated that the AI broadcast summary function through the SARSA service is used by an average of about 100,000 users per day.
"SOOP's AI services play a role in helping streamers run their content more easily and helping users find their preferences," said Lee Kyung-hwan, head of SOOP's AI Lab.
'Making Do Without' Voice Relay and Graphic Relay... Breaking Down Production Barriers
Beyond AI technology, SOOP is sparing no technical support to lower the barriers to content production for streamers. Representative examples are the "voice relay" and "graphic relay" systems unveiled to target the North and Central America World Cup period. Voice relay is a unique SOOP culture in which streamers commentate on game situations and communicate with users without directly broadcasting the game screen. The newly introduced graphic relay function visually displays real-time game data and records directly on the broadcast screen.
SOOP secured the relevant data and logo usage rights in advance, enhancing convenience so that streamers can produce high-quality broadcasts with the press of a button, without complex editing or production processes. This is the result of focusing on creating an environment streamers can actually use in the field, rather than on the flashiness of the technology itself.
"AI creates new interactions by reflecting a streamer's individuality, and graphic relay lowers the entry barrier to content production," a SOOP official said. "Based on communication data accumulated over years, SOOP will continue experiments to expand the unique value of live streaming even in the AI era."
