Whether it's a YouTube lecture, a TikTok, a podcast episode or a voice memo, there are four practical ways to turn speech into text. Here's when to use each.
1. YouTube's built-in transcript (free, captioned YouTube videos only)
On desktop, open the video, click ...more under the title, then Show transcript. It's good for skimming, but copying it out gives you a timestamp on every line.
2. yt-dlp (free, for developers)
yt-dlp --write-subs --write-auto-subs --sub-langs en --skip-download "<url>"
This saves the caption file without downloading the video. It works on many sites besides YouTube, but only if captions exist.
3. Whisper on your own machine (free, any audio)
When there are no captions, you need speech recognition. Download the audio (yt-dlp -x) and run faster-whisper or whisper.cpp. Use large-v3 for accuracy, and expect to need a GPU for long files.
4. A hosted tool (no setup)
If you don't want to install anything, paste the link or upload the file to a web tool. I built one, TranscribeLink: it uses captions when they exist and Whisper large-v3 otherwise, works with YouTube, TikTok, Instagram, X, Vimeo, SoundCloud and direct MP3/MP4 links, labels speakers, and exports TXT, DOCX, PDF, SRT and VTT. 20 minutes are free without an account. For recurring meetings, meeting-focused apps like Otter are a better fit.
Tips for better accuracy
- Clean audio beats a bigger model. Reduce background music and crosstalk when you can.
- Always check names, numbers and jargon before you publish a quote.
- For subtitles, export SRT or VTT so the timestamps stay attached to the text.
Which one should you use?
| You have | Use |
|---|---|
| A captioned YouTube video you just want to read | YouTube's Show transcript |
| Captions, and you script things | yt-dlp |
| No captions, and a GPU | Whisper locally |
| No captions, no setup, or a TikTok/Instagram/X link | A hosted tool |
Disclosure: I'm the developer of TranscribeLink.
Top comments (0)