Public runtime group for Vox Jot file transcription and speaker-isolation adapters that need Python dependencies beyond the live dictation engine.

This runtime covers Granite Speech, Cohere Transcribe, Higgs Audio v3 STT, PyAnnote, Reverb, WhisperX diarization, Polyvoice routing, and emotion2vec. Model weights stay in their original Hugging Face repos. Gated model repos still require publisher terms acceptance and a Hugging Face read token.

NeMo Sortformer is intentionally excluded because the available NeMo line requires vulnerable Transformers 4.x pins. Gemma audio, MLX ASR, and MLX Sortformer models use separate app-managed runtimes.

runtime-manifest.json is the app-facing manifest. requirements.txt pins the Python dependency set used by the managed runtime. Prebuilt platform archives can be uploaded as speech-analysis-runtime-<platform>.tar.gz.