I was originally using Parakeet for my speech recognition pipeline but decided to give Nemotron 3.5 a shot. After testing it on some multilingual audio clips, it's been working great so far. What sold me: - Better language support (40+ locales from one model) - Native streaming architecture — no more buffering entire files - Tested on CPU and got about 4.5x realtime speed using onnxruntime-genai as the backend I've containerized it with Docker so you can just clone and run. There are...