NVIDIA launches open-weight audio-native Nemotron models (2B, 30B)

The open-weight Nemotron models cover transcription, translation, sound recognition, audio Q&A, TTS, and full speech-to-speech. The speech stack is already quantized to GGUF for on-device use via NeMo-Speech.cpp, including Parakeet ASR models and Magpie-TTS Multilingual.