Separate narration creation from caption transcription

Record or obtain permissioned narration and retain the script, language and sample rate. A transcription tool reads speech; it does not create a narrator. faster-whisper documents CPU/GPU inference, word timestamps and VAD. Proposed workflow: transcribe the approved recording, compare words with the script, correct names and sentence boundaries, then review caption readability against the actual audio. Do not call word timestamps exact forced alignment. Chatterbox is an archival option for separately authorized research: the current README retirement banner says it is no longer maintained, despite stale service marketing lower down. Do not design a supported production dependency around that conflict. If evaluated later, review the exact model card, reference-voice consent, dependency pin and watermark behavior. Keep TTS source audio distinct from ASR transcript evidence. Keep a plain-text transcript and timed segments/words with start and end times. SRT or VTT subtitles need a formatting step that groups words into readable cues; word-timestamp output is not by itself a styled subtitle video. Choose CPU or a documented compatible GPU runtime from the pinned faster-whisper requirements. Compare proper nouns, pauses and overlapping speech with the source audio. Related existing Veyrul research: - [Does Chatterbox mark its generated audio, how can I check whether a clip carries that mark, and what does the code actually do?](https://veyrul.com/articles/art_01M5MEGCQWEXY3TGHFV0TPBBAY) (revision 1; reviewed for overlap). - [Where does the Chatterbox TTS licence appear: Hugging Face card prose, card metadata, or the GitHub repository?](https://veyrul.com/articles/art_01M5NPM0111J0X6S9CYN4NKYTN) (revision 1; reviewed for overlap). Resources: - [Transcription implementation](https://github.com/SYSTRAN/faster-whisper/blob/2a3a857737d17d974f30d818ebec69ce11929e5c/README.md) - [Archived TTS source and retirement caveat](https://github.com/resemble-ai/chatterbox/blob/ab55923c27757c90cfcc407baa57dc25a7d25217/README.md) Prerequisites: Permissioned audio; Model and environment approval. Cost: Local compute/storage/model transfer; no hosted fee assumed; zero calls executed. Limits: Review and correct ASR errors against the audio; Hardware fit and voice quality untested. Evidence: sources inspected on 2026-11-05 by founder-affiliated Veyrul researchers. Proposed steps have not been executed; no independent review, compatibility benchmark or outside-user result is claimed.

{
  "citations": [
    {
      "url": "https://github.com/SYSTRAN/faster-whisper/blob/2a3a857737d17d974f30d818ebec69ce11929e5c/README.md",
      "note": "faster-whisper pinned README"
    },
    {
      "url": "https://github.com/resemble-ai/chatterbox/blob/ab55923c27757c90cfcc407baa57dc25a7d25217/README.md",
      "note": "Chatterbox pinned README"
    },
    {
      "note": "Existing documentary background; no execution inferred.",
      "record_id": "art_01M5MEGCQWEXY3TGHFV0TPBBAY"
    },
    {
      "note": "Existing documentary background; no execution inferred.",
      "record_id": "art_01M5NPM0111J0X6S9CYN4NKYTN"
    }
  ]
}