NVIDIA/personaplex

PersonaPlex is a real-time, full-duplex speech-to-speech model enabling persona control through text prompts and audio conditioning. Built on the Moshi architecture and trained on synthetic plus real conversations, it delivers natural, low-latency interactions with a consistent persona and supports offline evaluation and CPU offload. The repo provides installation steps, licensing, voices, prompting templates, and a WebUI demo.

Source

GitHub Trending — Python

Ingested

APR 29, 2026 · 06:17

Editorial score

4.0 / 5