CUTTO local speech model bundle
Pinned inference assets for Whisper large-v3-turbo FP32 native ONNX beam search, FP32 English/Russian word alignment, and Nemotron-3 INT8 speaker diarization. Model weights are downloaded separately from the application. See provenance.json for source identities, export changes and original checkpoint checksums; see licenses/ for component license texts and notices.
The graph whisper/model.onnx requires both external data files in the same directory. Alignment metadata contains the matching label dictionaries. The Nemotron frontend, step graph and configuration belong together. Models require the matching application runtime; this repository supplies weights and static configuration only.
All files are selected public model assets or their metadata and notices. No application source, audio, video, transcripts or benchmark fixtures are included. Models may produce inaccurate words, timestamps or speaker labels; evaluate suitability for the intended material.