Improvement to quality of transcription. | Voters

Improvement to quality of transcription.

shipped

Steve Folland

It used to be pretty good, but it's getting worse compared to others. I have now switched to using ElevenLabs Speech to Text because it's 99.9% correct first time and then pasting in that transcription.

Anyway you can use their tool as your own transcription?

April 16, 2026

Gabe Michalski

marked this post as

shipped

We are marking this one as shipped as we just added ElevenLabs Scribe v2 as our default model for Transcription.
You may want to check your Settings → AI Models, as the previously labeled Scholar will still put you on a version of Rev.
Soon, we will have all of the options labeled correctly with the actual transcription provider.

Andy Kaufman

Gabe Michalski, this is great news. So we should select Scholar? Thanks!

Steve Folland

Gabe Michalski Excellent news! 🙌 I don't see it the AI Models for transcription yet but 🤞soon

Laura Burkhauser

Try again! We just updated our pipeline so this should now be DONE!

Gabe Michalski

marked this post as

under consideration

Transcription provider options are on the Roadmap but this might be out of scope for the 2 day telethon. Will keep you all posted.

Steve Folland

Thanks Gabe Michalski

Paul van der Mey

If there is cross-talk in multiple tracks, even when recorded in Rooms, it often confuses the speakers and swaps between them in apparently random ways while one person is speaking. Does it have no concept of the 'sound' of each voice, volume and delay differences to get that right?

Paul van der Mey

Even when a word is in my transcription glossary it can't reliably replicate it in the transcription.

Martin Morris

Absolutely agree. It is very tedious to make these small corrections

Mark Simon

Agreed with Andy. The transcripts could be a lot better. One constant issue beyond getting words right is that it treats any pause, even a slight one, as the end of a sentence. So not only are you fixing words, you're fixing how sentences read.

Andy Kaufman

This is easily the biggest headache of Descript. I waste so much time trying to make up for how bad the transcription is.