videodub-whisperx

VideoDub WhisperX ASR

FastAPI wrapper for WhisperX transcription/alignment used by the local VideoDub VideoLingo adapter.

Endpoints:

• GET /health
• POST /whisperx with multipart field audio_input plus optional form fields language, model, and align
• POST /align with multipart field audio_input, required segments_json, and optional language (default zh). This aligns a known transcript without loading or rerunning the Whisper transcription model.

After changing this Space, redeploy hf_asr_space/ before using videodub run --spoken-alignment whisperx. --spoken-alignment auto remains usable during rollout because it falls back to proportional timing when an old Space returns 404 for /align.

export HF_TOKEN=hf_xxx
hf upload lifengmds/videodub-whisperx hf_asr_space . \
  --repo-type space \
  --commit-message "Add transcript forced-alignment endpoint"

Set HF_TOKEN as a Space secret when private or gated Hugging Face models are needed.