link-transcript

name	link-transcript
description	Convert media links or local audio files into Chinese transcripts using the local link-transcript skill workflow and FunASR SenseVoiceSmall. Use when the user provides a Douyin/TikTok/Bilibili/social media URL, MP3, MP4, or audio/video file and asks to extract speech, transcribe, convert to text, or get the spoken content.
license	MIT. LICENSE.txt has complete terms.
compatibility	Requires Python 3.11+, ffmpeg, network access for URL parsing and first model download, and a Galaxy Downloader-compatible /api/parse backend.

Link Transcript

Use this skill to turn a media URL or local audio file into cleaned Chinese transcript text. Prefer the bundled script so the skill works after being installed as a standalone folder.

This skill and its repository are for learning and research only. Use it only with media the user owns, is licensed to process, or is otherwise authorized to process. Do not help bypass access controls, platform rules, copyright restrictions, or consent requirements.

Default Workflow

1. Locate the installed skill directory:
   • Prefer $LINK_TRANSCRIPT_SKILL_DIR when set.
   • Otherwise use the directory that contains this SKILL.md.
2. Ensure dependencies are available:
   • python3
   • ffmpeg
   • Python packages from requirements.txt
   • Install missing packages from the skill directory:

     python3 -m pip install -r requirements.txt
     

   • If setup is unclear, run:

     python3 scripts/check_install.py
     

   • On CPU-only Linux machines, install PyTorch CPU wheels first if a plain requirements install starts downloading large CUDA packages:

     python3 -m pip install --user --index-url https://download.pytorch.org/whl/cpu torch torchaudio
     python3 -m pip install --user -r requirements.txt
     

3. Ensure the downloader backend is available:
   • This skill delegates link parsing to a Galaxy Downloader-compatible API.
   • LINK_TRANSCRIPT_DOWNLOADER_API_BASE should point at a service exposing /api/parse.
   • Use the API base, not the Galaxy Downloader UI URL, unless the UI deployment also serves the compatible API route.
   • The default public endpoint is https://downloader-api.bhwa233.com, but for repeatable agent workflows prefer a deployment the user controls.
   • Read references/downloader.md when setup or response-shape details matter.
4. Run the bundled script:

   cd "<installed-link-transcript-skill-directory>"
   python3 scripts/transcribe_link.py "<url-or-audio-path>"
   

5. Return only the cleaned transcript text unless the user asks for files, logs, or implementation details.

Output Rules

• Do not summarize, polish, or rewrite the transcript.
• Remove SenseVoice tags such as <|zh|><|NEUTRAL|><|Speech|><|withitn|>.
• Mark 未人工校对 when recognition quality is uncertain or the text contains obvious ASR artifacts.
• Do not paste cookies, access tokens, or raw private request headers.

API Mode

API mode is optional and secondary. Read references/api.md before starting a reusable local HTTP service.

Configuration

• LINK_TRANSCRIPT_DOWNLOADER_API_BASE: downloader API base URL. Default is https://downloader-api.bhwa233.com.
• LINK_TRANSCRIPT_MODEL: default iic/SenseVoiceSmall.
• LINK_TRANSCRIPT_DEVICE: default cpu.
• LINK_TRANSCRIPT_LANGUAGE: default zh.

Acknowledgement

Downloader resolution is provided by Galaxy Downloader or a compatible deployment: https://github.com/lxw15337674/galaxy-downloader

Failure Handling

• If the downloader API fails, ask for a local MP3/MP4 file or a working Galaxy Downloader-compatible backend.
• If FunASR dependencies are missing or ffmpeg is unavailable, run python3 scripts/check_install.py and install from requirements.txt.
• If the first run is slow, explain that ModelScope is downloading iic/SenseVoiceSmall.

Reference Map

• references/downloader.md: downloader API contract and Galaxy Downloader setup notes.
• references/api.md: optional local FastAPI service mode.