At the 2025 World Artificial Intelligence Conference (WAIC), Soul App didn’t just show up—it made a statement. The social platform pulled back the curtain on its next-gen AI-powered social tools, headlined by a proprietary full-duplex voice large language model (LLM) that promises to overhaul how humans and AI interact in real time.
Forget awkward pauses and robotic turn-taking—Soul’s new model is designed to talk like you do, think on the fly, and even moderate your group chats.
From Voice Bots to Real Conversations
In a live demonstration titled “Audio Partyroom,” Soul showcased an AI host moderating multi-user voice chats with a level of fluency and emotional intelligence that could make even seasoned human moderators sweat.
This wasn’t just a technical flex. The full-duplex model ditches the traditional Voice Activity Detection (VAD) and rigid latency logic that most voice AIs rely on. Instead, it brings autonomous rhythm control, creating real-time, reciprocal voice interactions where both humans and AI can speak—and be understood—naturally.
The model will enter beta testing soon, with applications across one-on-one matching, real-time group chats, and even AI-human calls in virtual spaces. That last part? Yes, AI can now host your next digital dinner party.
Emotional Intelligence: Soul’s Secret Weapon
What separates Soul from your average social app isn’t just voice tech—it’s emotional intelligence at scale. CTO Tao Ming calls this a “revolutionary paradigm” where AI doesn’t just respond but connects—bridging emotional gaps and acting as both support system and matchmaker.
“Soul is cultivating a new dimension of digital belonging—one that delivers genuine joy and transforms how we perceive community,” said Ming.
With AI-assisted emotional mapping, contextual awareness, and interest-based user matching, Soul’s LLM doesn’t just filter content—it reads the room.
Next Stop: Multimodal AI Interaction
Voice is just one piece of Soul’s AI puzzle. The company is pushing into real-time video, combining audio with dynamic portrait video generation, allowing users to “see” AI personas in real time. This multimodal experience is part of Soul’s broader effort to humanize digital interaction, blurring the lines between virtual and real.
The research behind this effort just earned Soul a spot at CVPR 2025, one of the world’s top academic AI conferences—underscoring its credibility beyond hype.
More Than a Social App: A Playground for AI Companionship
Founded in 2016, Soul has long marketed itself as a low-pressure space for Gen Z to connect around shared interests. But in recent years, the platform has evolved into an AI-driven emotional ecosystem, leveraging public social data and in-house R&D to continuously iterate its core products.
By 2024, it had consolidated its work into a comprehensive multimodal AI system—covering voice, text, digital avatars, and music generation. Its Soul X model underpins everything from intelligent chat to real-time emotional response engines.
And it’s not stopping there. Soul is developing a network of Virtual Influencers and AI-assisted content tools to lower expressive barriers, helping users navigate social anxiety, workplace drama, or just the challenge of saying “hi” to someone new.
Riding the AI Social Wave
Soul’s strategy aligns with a broader trend in 2025: the explosion of “AI+” applications. As AI becomes more deeply integrated into everyday tools—from productivity to companionship—Soul’s bet on emotional-value commerce feels well-timed.
Unlike traditional platforms that mine engagement through endless scrolls, Soul is attempting to cultivate authentic emotional experiences powered by AI, and tailored for a demographic that increasingly values mental wellness, real connection, and authentic expression.
This is more than a UI refresh. Soul is reengineering the why of social networking—and if its WAIC showcase is any indicator, the platform sees AI not just as an assistant, but as an essential cohabitant in our digital lives.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI.
