NVIDIA has quietly released a new speech recognition model that could be a game-changer for local voice processing. The model, called Nemotron-3.5-ASR, is a 0.6 billion parameter open-source model designed specifically for real-time streaming. Its performance is impressive enough to potentially shift the landscape for on-device voice pipelines. The key highlights include its efficiency and accuracy, as demonstrated in benchmarks. More details can be found in the linked thread.