NVIDIA published Nemotron 3.5 Transcribe, an open-weights speech model that can change transcript style based on natural-language instructions.
It supports verbatim or written-form output, speaker identification, translation and language ID across about 40 languages, plus hour-long recordings with word-level timestamps. Users can also supply expected names and terminology to reduce domain errors.
Open speech models like this are becoming building blocks for support agents, meeting bots and multilingual marketing workflows.
Questions? Chat with our team — threaded comments are coming soon.