NVIDIA Open-Sources PersonaPlex-7B, a Free Full-Duplex Voice AI That Listens and Speaks at Once

NVIDIA has released PersonaPlex-7B, a free, open-source speech-to-speech model that enables simultaneous listening and speaking—a milestone for conversational AI that until now has required cascading separate systems. Released January 15, 2026, the 7-billion-parameter model collapses the traditional architecture of automatic speech recognition (ASR), large language model (LLM), and text-to-speech (TTS) into a single Transformer that ingests and outputs audio tokens in real time, eliminating the rigid turn-taking behavior that has defined voice assistants since Siri.