video-Zebra-china

Codex skill and standalone scripts for turning YouTube or local foreign-language videos into:

• Simplified Chinese subtitles
• Mandarin dubbing
• final MP4 exports with burned-in subtitles

The current pipeline uses:

• yt-dlp for YouTube download
• faster-whisper for local ASR
• Volcengine TranslateText for subtitle translation
• Volc TTS openspeech.bytedance.com/api/v3/tts/unidirectional for Mandarin dubbing
• ffmpeg for audio mixing and final export

Repository Layout

• SKILL.md: Codex skill entry
• scripts/localize_video.py: end-to-end pipeline
• scripts/run_localize_video.sh: launcher that prefers .venv and falls back to python3
• references/runtime-requirements.md: setup notes and troubleshooting
• agents/openai.yaml: skill metadata

Quick Start

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
brew install ffmpeg yt-dlp

Set credentials:

export VOLCENGINE_ACCESS_KEY="AK..."
export VOLCENGINE_SECRET_KEY="SK..."
export VOLCENGINE_REGION="cn-north-1"
export VOLCENGINE_TTS_API_KEY="your-tts-api-key"
export VOLCENGINE_TTS_RESOURCE_ID="volc.service_type.10029"

Run with a YouTube URL:

scripts/run_localize_video.sh \
  --input "https://www.youtube.com/watch?v=..." \
  --workdir ./runs/demo

If YouTube blocks anonymous download, reuse browser cookies:

scripts/run_localize_video.sh \
  --input "https://www.youtube.com/watch?v=..." \
  --workdir ./runs/demo \
  --cookies-from-browser edge

Run with a local file:

scripts/run_localize_video.sh \
  --input /path/to/video.mp4 \
  --workdir ./runs/local-video \
  --voice zh_male_beijingxiaoye_emo_v2_mars_bigtts

Outputs

The pipeline writes these files into the selected work directory:

• source.*
• audio.wav
• transcript.json
• translated_segments.json
• subtitles.zh.srt
• dub_track.wav
• final_audio.wav
• final.mp4