From Startup Fortune · · 1m
Nvidia Releases Real-Time Voice AI Model
Nvidia's open-source voice AI handles conversations without waiting for replies.
In brief
Nvidia's open-source voice AI handles conversations without waiting for replies. Nvidia released PersonaPlex-7B, an open-source voice model that listens and speaks simultaneously, reducing response delays to 240 milliseconds. Originally reported by Startup Fortune.
Nvidia Releases Voice AI Model
Nvidia released PersonaPlex-7B, a free open-source voice model that handles conversations without waiting for replies.
Real Conversation Flow
Real conversation involves interruptions and overlapping speech, which traditional voice AI systems cannot handle smoothly.
Release Date and Details
Nvidia released PersonaPlex-7B-v1 on January 15, 2026, with MIT license and NVIDIA Open Model License.
Full-Duplex Technology
PersonaPlex uses dual-stream design to listen and speak simultaneously, eliminating handoffs between systems.
Performance Metrics
The model achieves 90.8% turn-taking rate and interruption response latency as low as 240 milliseconds.
Control Layer Functionality
PersonaPlex takes voice and text prompts to define persona and vocal characteristics for different roles.
Industry Impact
Nvidia's open-source model sets a public benchmark for real-time speech latency and turn-taking accuracy.
The bottom line
Nvidia released PersonaPlex-7B, an open-source voice model that listens and speaks simultaneously, reducing response delays to 240 milliseconds.
