twat-speech

twat-speech is the speech-domain package for the twat ecosystem. It owns transcription orchestration, subtitle repair, TTS calls, and deterministic dubbing plans while leaving provider clients in twat_genai and audio-file transforms in twat_audio.

Python API

from twat_speech import repair_subtitles, dub_plan, transcribe, synthesize

fixed = repair_subtitles(open("captions.srt").read())
plan = dub_plan("movie.mp4", fixed, language="de")
transcript = transcribe("speech.wav")
speech_file = synthesize("Hello", "hello.wav")

Implemented helpers:

• parse_srt, format_srt, repair_srt, repair_subtitles.
• subtitle_transcript for plain transcript extraction.
• transcribe and synthesize as thin twat_genai provider adapters.
• dub_plan / create_dub_plan for deterministic dubbing orchestration.
• optional transcribe(..., preprocess=True) delegates resampling to twat_audio before provider transcription.

CLI

python -m twat_speech --help
python -m twat_speech repair-srt captions.srt captions.fixed.srt
python -m twat_speech dub movie.mp4 captions.fixed.srt --language de
python -m twat_speech transcribe speech.wav
python -m twat_speech synthesize "Hello" hello.wav

Boundaries

• twat_genai owns speech providers for STT/TTS and future dubbing engines.
• twat_audio owns audio-file transforms such as resampling, trimming, normalization, and audio extraction.
• twat_speech owns speech-specific orchestration, subtitles, transcripts, and dubbing plans.

Reference scripts

Reusable subtitle repair ideas from srtmultifix.py, extraction workflows from mkv2srt.py, and planning ideas from vidpredub.py/viddub-* are represented here as package APIs. Provider-specific or workstation-specific batch dubbing scripts remain in reference/bin-img-vid/ because they combine local credentials, shell glue, and one-off media workflows.