Captions first.
Read CORS-accessible captions or page subtitle tracks and skip inference. If a platform restricts access, the app explicitly offers tab-audio consent instead.
LESS WATCHING. MORE UNDERSTANDING.
Turn videos
into articles.
Paste a link. Read captions first.
When captions are unavailable,
let your browser listen.
Browser-only · No installation · GPU optional, CPU supported
One link. A readable article.
Keep the source. Let your browser transcribe, and your chosen model organize the ideas.
No readable captions? Listen.
Explicitly share the video tab audio. Whisper runs in a browser Worker while you play it. No cloud ASR, extension or local service required.
Edit with evidence.
Use your API to generate four article depths. Correct clear transcription errors, preserve numbers and qualifications, and flag uncertain terms while retaining the transcript.
One source. Four reading depths.
Length follows information density. No invented filler or silently dropped ending.
Short200–350 words
The core conclusion
The core conclusion
Medium600–1,000 words
Main ideas and evidence
Main ideas and evidence
Long1,200–2,000 words
Examples and reasoning
Examples and reasoning
Detailed2,000–3,500 words
Processes and limits
Processes and limits
YOUR BROWSER. YOUR COMPUTE.
No GPU? Still works.
This screenshot comes from real Chrome: Whisper recognized a public speech sample using CPU / WASM, without an ASR backend.
Try WebGPU when available, otherwise run the quantized model on CPU. Audio capture and inference use separate workers, with visible pending time. Model assets download on demand and can be cached.
Read the verification record ↗
Open the website. Start reading.
No Python, yt-dlp or local model service. Desktop Chrome / Edge is the initial recommended browser for tab-audio sharing.
-
Paste a link. Check captions.
Choose an article depth. The app first tries accessible captions.
-
Share tab audio if needed.
Use the inline player and explicitly share this tab audio. Restricted embeds can fall back to the original video tab.
-
Finish transcription. Get an article.
Stop capture, drain the queue, then generate with your API. Copy or export the full text.
Audio stays in the browser. Only requested generation sends text to your chosen API. Keys remain in page memory.
A few things to know.
Is this website the actual app?
Yes. Open the app and use it without a local service. The current app interface is Chinese, with Chinese and English generation and bilingual product pages.
Why share the video tab audio?
Webpages cannot freely read protected audio or captions from other sites. Consent shares playback, not a downloadable entire audio track. Capture follows playback; pauses, seeking, ads and speed changes affect timestamps. Skipped content is not recovered automatically.
What if I have no GPU?
Use the lightweight Whisper tiny model through CPU / WASM. Speed depends on your device. Pending time is visible, and processing can continue after capture ends. GPU acceleration is optional.
Are ASR and article generation free?
Browser ASR does not call a paid transcription service. Initial use downloads an open model. Article generation uses your selected API; its charges or free tier depend on the provider. Custom services must allow browser CORS; no shared key is bundled.
What happens if recognition fails?
Earlier text and queued audio are retained. Retry on CPU or download pending audio. Partial results are labelled. Editorial correction is not external fact checking; important claims still need verification.