NVIDIA's Real-Time Full-Duplex Voice Model

Back to Devoured - August 05, 2026

DEVOURED

NVIDIA's Real-Time Full-Duplex Voice Model

NVIDIA released NemotronLabs VoiceChat, an 11B parameter end-to-end speech model capable of real-time full-duplex communication and tool calling.

What: The model architecture integrates speech understanding, generation, and tool invocation within a single 11B parameter neural network to handle streaming voice interactions.

Why it matters: The industry is consolidating voice-to-voice interaction from multi-stage pipelines (ASR to LLM to TTS) into monolithic, end-to-end models to reduce latency in conversational agents.

Decoder

Full-duplex: A communication system that allows for simultaneous transmission and reception of data, enabling natural interruptions and 'barge-in' behavior in speech models.

Original article

NemotronLabs VoiceChat is an 11B end-to-end speech model that handles streaming understanding, speech generation, and tool calling within one architecture.

Explore all 55 articles in today's digest