Deepdub Launches Lightning, a Real-Time Voice Model for Scalable AI Agents and Enterprise Applications with Ultra-Low Latency

Built on Deepdub’s Foundational Voice AI Models and NVIDIA accelerated computing, Lightning 2.5 delivers 2.8X more throughput and 5X higher concurrency than the previous version, as well as latency as low as 200 milliseconds, half a second faster than industry standards

Deepdub, a foundational voice AI model company pioneering expressive localization technologies, announced the release of Lightning 2.5, its fastest and most scalable voice model yet. Built for real-time, production-grade voice experiences, the new model delivers exceptional support for multilingual AI agents, contact center automation, real-time content localization, and personalized media and gaming, among other applications.

Developed on Deepdub’s proprietary language and speech foundation models, Lightning combines human-level expressiveness and high speaker fidelity across multiple languages, with infrastructure scaling for real-time performance. It preserves speaker identity, intonation, and emotional nuance, making it perfectly suited for lifelike, production-quality voice experiences.

Marketing Technology News: MarTech Interview with Haley Trost, Group Product Marketing Manager @ Braze

“Lightning represents a leap forward in real-time, expressive voice generation,” said Ofir Krakowski, CEO and Co-founder of Deepdub. “Whether powering multilingual AI agents, immersive content platforms, or voice-first automation at scale, Lightning brings foundational voice AI to production with the speed, fidelity, and quality our customers demand.”

Lightning 2.5 introduces a redesigned inference engine that delivers 2.8X more throughput and 5X higher concurrency than the previous version. With latency as low as 200 milliseconds, half a second faster than industry standards, Lightning enables real-time responsiveness in workflows such as live voice interactions, dynamic voiceovers, and event-driven AI pipelines. This makes Lightning ideal for use in autonomous voice agents, interactive systems, and large-scale voice automation across industries.

A member of the NVIDIA Inception program for startups, Deepdub has optimized Lightning for deployment across a range of NVIDIA GPUs, including NVIDIA H100, NVIDIA A100, NVIDIA L40S, and NVIDIA A10G, with support for both on-premises and cloud environments. Leveraging NVIDIA TensorRT-LLM acceleration stack, Lightning delivers high-throughput, low-latency inference at enterprise scale, while maintaining efficient operational cost.

Marketing Technology News: Cross-Department Collaboration with Marketing Workflow Automation: Enhancing Alignment Between Sales, Customer Service, and Marketing Teams

“Organizations are demanding real-time, expressive voice AI that combines lifelike quality with speed and scalability,” said Richard Kerris, Vice President of Media and Entertainment at NVIDIA. “By building Lightning 2.5 on NVIDIA GPUs and TensorRT-LLM, Deepdub enables ultra-low latency and high concurrency for multilingual voice experiences, helping customers deliver next-generation AI agents, immersive content, and enterprise voice applications at global scale.”

Deepdub will showcase Lightning at IBC 2025, featuring live demos of real-time voice localization, dynamic content automation, and scalable voice pipelines for media and broadcast platforms. Visit us at the AI Pavilion, Hall 14, Booth 14.B53 to experience Lightning in action.