NVIDIA PersonaPlex-7B: Full-Duplex Voice AI Model Trending on HuggingFace With 500K Downloads
GitHub·medium signal
NVIDIA's PersonaPlex-7B is a real-time, full-duplex speech-to-speech conversational model built on a 7B transformer backbone with Mimi audio codecs at 24kHz. It bypasses the traditional ASR→LLM→TTS pipeline entirely, handling interruptions naturally while maintaining persona control through text prompts and audio conditioning. MIT-licensed code and NVIDIA Open Model license for weights. Nearly 500K HuggingFace downloads and 9.7K GitHub stars, ranking among the top 6 open-source AI models of April 2026.