NVIDIA Unveils Full-Duplex Voice Conversation Model: PersonaPlex-7B-v1
2026-01-20 / Read about 0 minute
Author:小编   

NVIDIA Research has just launched its groundbreaking full-duplex voice conversation model, PersonaPlex-7B-v1, which boasts an impressive 7 billion parameters. Built upon the Moshi architecture, this innovative model leverages a unified Transformer framework, enabling it to seamlessly handle natural interruptions and overlapping speech—features that closely mimic real-life human conversations. In rigorous evaluations, PersonaPlex-7B-v1 has showcased outstanding performance, marking a significant leap forward in voice interaction technology. To foster further innovation and collaboration within the community, NVIDIA has generously open-sourced the code, with model weights released under the NVIDIA Open Model License.