Congratulations on the AI Transcription implementation!

I just completed my first transcription in DEVONthink 4 and the timestamps with links into the original audio from the annotation file is a first for me. Really great. I use a whisper based transcription tool and I think the quality of the transcription itself may be better than the Apple Speech models, but those links may sway which tool I use in practice. Being able to edit the transcriptions right in the annotation file and still be able to link directly for clarification is really great.


3 Likes

Thank you for the nice feedback, we’re glad that you like this new feature :slight_smile:

1 Like

omg, I am very very new to DEVONthink-- but THIS IS WHAT I WANT, how did you do this? I wrote a book with more than 200 interviews all stored in otter and I need to get it off of that service and into my own computer. can you share how to do this?

First set up the desired service & destination in Settings > AI > Transcription, afterwards transcription can be applied e.g. via the menu item Data > Recognition > Transcribe Speech to audio (e.g. MP3 or MP4) or video files.

this is so exciting… however, after trying and configuring it… I am not getting timestamps and /or formatting. I am using the chat gpt version w/ api key.

  • Are you running beta 3, released today?
    • If so, did you enable AI > Transcription > Add timestamps to transcription ?

using Devon4 the beta

Correct. Did you enable that option in the DEVONthink > Settings?

I have 4.0.1 and confirm that timestamp does not work since beta 3. And yes, it is enabled in settings.

A complete screenshot of the settings would be useful.

sorry for late reply. Here it is

Transcribing to (usually invisible) searchable text doesn’t include timestamps.

I understand, but I was using “Insert Transcription” feature in the annotations inspector (screenshot attached). Therefore, I believe it would be better if it respected my timestamp settings.

But my problem solved. Thanks.

If the output of the transcription is searchable text, no timestamps are being created. So inserting the transcript wouldn’t have timestamps either. They’re not added after the fact.

Also, if you’re using the transcript in an Annotation file, why not set that as the Destination in the AI > Transcription settings?

Does the time stamping only work with the GPT-4o transcription? I tried it with the online Apple transcription option with time stamping checked, but there was no time stamping in the result.

Only the transcription to comments or annotations uses time stamps.

Thanks for the quick reply, and sorry for my ignorance. This is an exciting feature which I‘d love to use fully. So if I have an mp3 file, what do I have to do to have it transcribed with time stamps?
I used the context menu and Erkennung > Sprache transkribieren. What should I do instead?

In Settings > AI > Transcription select the destination Comment or Annotation first.

Thanks - marvellous! I would have killed for that feature when I had to manually transcribe more than a dozen interviews twenty years ago… Great!
Keep up the good work!

The Meeting Notes templates in the menu Data > New from Template > AI might be useful too.

1 Like