Most free transcription tools quietly cap you at ~15 minutes, or hold your transcript hostage behind a signup. I wanted the opposite: paste a link to a two-hour podcast or a full lecture and just get the text — free, no account. That became Longscribe. Here’s what turned out to be harder than expected.
1. Long audio breaks the naive approach
Feeding a 2-hour file straight into Whisper is a great way to time out. The fix was chunking on silence boundaries, transcribing chunks with bounded retries, and stitching timestamps back together so the final SRT subtitles still line up.
2. Captions-first beats brute force
For YouTube transcription, downloading and transcribing audio is the expensive path. If a video already has captions, use them — it’s faster, cheaper, and far kinder to rate limits. Only fall back to audio + Whisper when there genuinely are none.
3. “Free” means every request has to be cheap
No paid APIs anywhere. Jobs run on a background queue so you can close the tab and get an email when it’s done, instead of holding a browser connection open for 40 minutes.
4. Export formats are the actual product
People don’t want a text blob — they want timestamped segments, .srt subtitles, and a .docx they can hand to someone. That last mile mattered more than the model.
It’s live and free at longscribe.com if you want to throw a long recording at it.
답글 남기기