PersonaPlex

weights

open(7B safetensors checkpoint on the Hub behind an automatically approved license click-through)

data

closed(Fisher English (an LDC corpus) plus synthetic conversations generated with open LLMs and TTS

code

partial(inference and server code only)

license

NVIDIA-Open-Model-License(the weights

The weights are published under the NVIDIA Open Model License, which permits commercial use and derivatives, and the code is MIT. NVIDIA trained it on the licensed Fisher corpus plus synthetic conversations it has not released, and publishes inference code only.

• https://huggingface.co/api/models/nvidia/personaplex-7b-v1 recorded 2026-09-27

  Hub record: license_name "nvidia-open-model-license", "gated":"auto", and model.safetensors in the file list.

• https://raw.githubusercontent.com/NVIDIA/personaplex/main/README.md recorded 2026-09-27

  README: "The present code is provided under the MIT license. The weights for the models are released under the NVIDIA Open Model license."; sections "Launch Server" and "Offline Evaluation".

• https://research.nvidia.com/labs/adlr/personaplex/ recorded 2026-09-27

  Project page: "PersonaPlex trains on 7,303 real conversations (1217 hours) from the Fisher English corpus"; "The conversation speech was generated using Chatterbox TTS".

• https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ recorded 2026-09-27

  NVIDIA Open Model License Agreement: "Models are commercially usable. You are free to create and distribute Derivative Models."

Hugging Face downloads of the single released checkpoint.

• https://huggingface.co/api/models/nvidia/personaplex-7b-v1 recorded 2026-09-27

  "downloads":194423 for nvidia/personaplex-7b-v1.

Holds a live two-way spoken conversation like Moshi, on which it is built, and adds control over who the assistant is and how it sounds.

• https://huggingface.co/nvidia/personaplex-7b-v1 recorded 2026-09-27

  Card: "Personaplex runs in a dual-stream configuration in which listening and speaking occur concurrently."; "Testing/Evaluation Dataset: Link: FullDuplexBench".

• https://raw.githubusercontent.com/NVIDIA/personaplex/main/README.md recorded 2026-09-27

  README: "PersonaPlex is a real-time, full-duplex speech-to-speech conversational model that enables persona control through text-based role prompts and audio-based voice conditioning."