The company describes a single open, end-to-end checkpoint that labels up to eight speakers and offers algorithmic latency profiles from 0.32 to 30.4 seconds.
Baseten says its RTX PRO 6000 deployment sustained more than 500 concurrent hour-long diarization streams, or 190 with transcription added. The release also positions per-speaker activity as an input for voice activity detection, endpointing, interruption handling, and speaker-attributed transcription with NVIDIA Parakeet V2 ASR.