Deepdub Launches Lightning: Real-Time Voice Model for AI Agents

Built on Deepdub’s Foundational Voice AI Models and NVIDIA accelerated computing, Lightning 2.5 delivers 2.8X extra throughput and 5X greater concurrency than the earlier model, in addition to latency as little as 200 milliseconds, half a second quicker than business requirements

Deepdub, a foundational voice AI mannequin firm pioneering expressive localization applied sciences, introduced at the moment the discharge of Lightning 2.5, its quickest and most scalable voice mannequin but. Built for real-time, production-grade voice experiences, the brand new mannequin delivers distinctive help for multilingual AI brokers, contact heart automation, real-time content material localization, and personalised media and gaming, amongst different purposes.

Developed on Deepdub’s proprietary language and speech basis fashions, Lightning combines human-level expressiveness and excessive speaker constancy throughout a number of languages, with infrastructure scaling for real-time efficiency. It preserves speaker id, intonation, and emotional nuance, making it completely suited for lifelike, production-quality voice experiences.

“Lightning represents a leap ahead in real-time, expressive voice technology,” mentioned Ofir Krakowski, CEO and Co-founder of Deepdub. “Whether powering multilingual AI brokers, immersive content material platforms, or voice-first automation at scale, Lightning brings foundational voice AI to manufacturing with the pace, constancy, and high quality our clients demand.”

Lightning 2.5 introduces a redesigned inference engine that delivers 2.8X extra throughput and 5X greater concurrency than the earlier model. With latency as little as 200 milliseconds, half a second quicker than business requirements, Lightning allows real-time responsiveness in workflows reminiscent of reside voice interactions, dynamic voiceovers, and event-driven AI pipelines. This makes Lightning best for use in autonomous voice brokers, interactive programs, and large-scale voice automation throughout industries.

A member of the NVIDIA Inception program for startups, Deepdub has optimized Lightning for deployment throughout a spread of NVIDIA GPUs, together with NVIDIA H100, NVIDIA A100, NVIDIA L40S, and NVIDIA A10G, with help for each on-premises and cloud environments. Leveraging NVIDIA TensorRT-LLM acceleration stack, Lightning delivers high-throughput, low-latency inference at enterprise scale, whereas sustaining environment friendly operational value.

“Organizations are demanding real-time, expressive voice AI that mixes lifelike high quality with pace and scalability,” mentioned Richard Kerris, Vice President of Media and Entertainment at NVIDIA. “By constructing Lightning 2.5 on NVIDIA GPUs and TensorRT-LLM, Deepdub allows ultra-low latency and excessive concurrency for multilingual voice experiences, serving to clients ship next-generation AI brokers, immersive content material, and enterprise voice purposes at international scale.”

Deepdub will showcase Lightning at IBC 2025, that includes reside demos of real-time voice localization, dynamic content material automation, and scalable voice pipelines for media and broadcast platforms. Visit us on the AI Pavilion, Hall 14, Booth 14.B53 to expertise Lightning in motion.

The publish Deepdub Launches Lightning: Real-Time Voice Model for AI Agents first appeared on AI-Tech Park.