NemotronLabs VoiceChat 11B

model
NVIDIA

End-to-end full-duplex speech-to-speech model: FastConformer encoder + Nemotron Nano 9B v2 hybrid-Mamba backbone + TTS decoder, with ~450ms turn-taking, barge-in, and a claimed first for an open model — live mid-conversation tool calling. #2 on VoiceBench and Full-Duplex-Bench at release. OpenMDW 1.1 license.

HuggingFace

Model Details

Architecture DENSE
Parameters 11B
License OpenMDW 1.1
audioopen-weightagents

Related