ytdlp-asr-funasr

name	ytdlp-asr-funasr
description	Download audio from a video URL with yt-dlp and transcribe it with FunASR. Use when the user shares a YouTube/Bilibili/other yt-dlp-supported video link and wants the spoken content, transcript, summary, subtitles, or a "watch this video" style analysis based on actual audio instead of page text.

yt-dlp + FunASR transcript

Turn a video URL into text: download audio with yt-dlp, transcribe it with FunASR, then summarize or answer questions from the transcript.

Workflow

1. Run scripts/url_to_transcript.py with the target URL.
2. Read the generated *.metadata.json first to confirm title, paths, duration, and whether long-audio mode was used.
3. Read *.transcript.txt and use that text for summaries, Q&A, timestamps, and note extraction.
4. If the user needs visual details, use a separate video-frame workflow; do not pretend an audio-only transcript includes visual analysis.

Quick usage

uv run {skillDir}/scripts/url_to_transcript.py "https://www.youtube.com/watch?v=..."

Common options:

# Force language
uv run {skillDir}/scripts/url_to_transcript.py URL -l zh

# Custom output directory
uv run {skillDir}/scripts/url_to_transcript.py URL -o ./output

# Adjust long-audio behavior
uv run {skillDir}/scripts/url_to_transcript.py URL --long-threshold 180 --chunk-size 20

Output files

Default output directory: {skillDir}/output/

• TIMESTAMP-VIDEOID.wav: extracted audio
• TIMESTAMP-VIDEOID.transcript.txt: transcript text
• TIMESTAMP-VIDEOID.metadata.json: metadata (title, duration, paths, long mode)

Always inspect metadata before picking the transcript file.

Environment

Required external commands:

• uv
• yt-dlp
• ffmpeg
• ffprobe

Python dependencies are managed by pyproject.toml and installed automatically by uv run.

Runtime details

• yt-dlp --dump-single-json is used first to get title and video ID.
• Output names use timestamp + video_id to avoid collisions.
• Audio is downloaded as WAV for reliable transcription.
• Long audio automatically adds --long when duration exceeds --long-threshold.
• Transcription is handled by scripts/transcribe.py in the same skill.

Failure handling

yt-dlp fails

Check whether yt-dlp exists:

command -v yt-dlp

If the site is throttling or requires a newer extractor, inspect stderr and update yt-dlp.

FunASR import fails

Install dependencies inside the skill directory:

uv sync

Then retry:

uv run {skillDir}/scripts/transcribe.py AUDIO.wav -o out.txt

Output too long

Prefer summarizing the transcript instead of pasting the whole file. Quote only the relevant passages when needed.