NemotronLabs VoiceChat introduces a single open model that combines full-duplex voice interaction with native tool-calling capabilities. It uses a streaming speech encoder and a decoder-only language model with parallel output streams for agent text and structured function calls, an auxiliary RNN-T branch for incremental user transcription, and a streaming TTS decoder.
On Full-Duplex-Bench 1.0, the model achieved the lowest pause-handling takeover rates among open-weight systems, perfect interruption handling, and high post-interruption response quality. It also showed strong backchannel handling and a 82.5% tool-selection F1 score. This work is interesting because it integrates real-time conversational dynamics with tool use in a single open architecture, eliminating the need for cascaded systems.