Transcribe audio and video files via Alibaba Cloud Tingwu REST API into Markdown.
npx skills add https://github.com/cat-xierluo/legal-skills --skill tingwu-asr
Please help me install this Agent Skill. Skill: tingwu-asr Source: https://github.com/cat-xierluo/legal-skills/tree/main/skills/tingwu-asr Command: npx skills add https://github.com/cat-xierluo/legal-skills --skill tingwu-asr
Local speech-to-text tools like FunASR may be unavailable, too slow, or inaccurate for long recordings. This Skill offloads transcription to Alibaba Cloud Tingwu, turning audio/video files or podcast links into structured Markdown transcripts with speaker separation, timestamps, and AI-generated summaries.
Ask the AI to transcribe your audio or video file with Tingwu, for example: "Transcribe /path/to/meeting.mp4 with speaker separation using tingwu-asr."
High-intent search queries and answers about installing and using this skill.
Run the transcribe.py script with your file path, for example python3 scripts/transcribe.py audio.mp3 --lang cn. The script uploads the file to Alibaba OSS, starts cloud transcription, polls until done, and writes a Markdown transcript with timestamps next to the source file and in an archive directory.
Pass the URL directly to transcribe.py; it uses yt-dlp to download the audio from sources like Xiaoyuzhou, YouTube, or Bilibili, then uploads it for cloud transcription. This local-upload path is required because Tingwu's native podcast-link channel disables speaker separation.
No. Testing shows Tingwu's roleSplitNum only works with the value 2, which separates two speakers; values 0, 1, 3, and 4 produce no separation. The skill defaults to --speakers 2 for reliable two-person diarization.
Supported formats include mp3, wav, m4a, wma, aac, ogg, flac for audio and mp4, mov, mkv, webm, avi for video. Audio files are limited to 500MB, video to 6GB, and a single file can be up to 6 hours long.
Tingwu authentication relies on cookies that expire over time. Re-run the built-in Playwright script login_pw.py, which opens a browser, fills credentials from config/.env, and saves fresh cookies including the HttpOnly login ticket directly to config/cookies.json.
Yes. Use the --async flag to upload and submit the task, which returns immediately with a task ID saved to pending_tasks.json. Then run poll_tasks.py --monitor in the background to poll status and automatically generate the Markdown output when finished.
taracodlabsExtract transcripts, download audio and video, and retrieve metadata from YouTube videos.
takusaotomeGenerate meeting transcripts and structured Markdown minutes from video files.
linuxhsjTranscribe and translate audio files locally using the Whisper CLI.
bitan-delTranscribe audio files locally using the Whisper CLI.
