Related News
Oct 29, 2026
Granary-v2: what is in NVIDIA’s public speech-label release
Granary-v2 releases labels for 463,062 hours across 48 languages. Its larger processed collection, source audio and per-record targets have separate scopes and requirements.
Oct 28, 2026
Nemotron 3.5 Transcribe handles recordings, with task and language limits
NVIDIA’s new open-weight audio model supports transcripts, translations and questions about recordings. Offline processing, language coverage and speaker labels each have a distinct boundary.
Oct 28, 2026
LangSmith adopts reference_outputs for creating dataset examples
LangSmith’s October 19–26 updates name expected example outputs reference_outputs. The old outputs alias remains valid, but sending both is rejected.