cjm-capability-voxtral-hf

A Mistral Voxtral speech-to-text capability for the cjm-substrate runtime that provides local transcription through Hugging Face Transformers with configurable model selection and parameter control.

Modules

• cjm_capability_voxtral_hf.capability — Capability implementation for Mistral Voxtral transcription through Hugging Face Transformers (Option C, stage 8: pure compute).
• examples/hf-testing.ipynb

API

cjm_capability_voxtral_hf.capability

• VoxtralHFCapability class — Mistral Voxtral transcription capability via Hugging Face Transformers (stage 8: pure-compute tool capability).
• VoxtralHFCapabilityConfig class — Configuration for Voxtral HF transcription capability.

Dependencies

Depends on: accelerate, cjm-capability-primitives, cjm-substrate, cjm-substrate-hf-utils, cjm-substrate-torch-utils, fastcore, librosa, mistral-common, numpy, soundfile, torch, transformers