I’ve been experimenting with using Tana’s meeting transcription service, which is not just for meetings. It’s for almost any streaming audio on the local computer. So, I can use it to transcribe podcasts, conference presentations, etc. (Note: this is for my personal use, I don’t send these results to anyone.)
Tana’s service mis-hears non-English phrases, locations, and uncommon words. So, I export the transcripts from Tana to HTML then load the raw transcript into Claude. I tell Claude who the speakers are (Tana doesn’t identify speakers), tell it to make well-formed paragraphs from the conversation, clean up misspellings, and identify the speakers in the text. It does a very good job of that.
Sometimes, from that, I then put the transcript into Tinderbox for my own note taking, linking, etc. Claude helps further by posing challenges to positions or statements from the speaker(s), adding useful facts, mapping out the point-counterpoint dialog, etc.
There’s nothing like a good integration. Especially in a case where AI is doing useful things to advance the analyses and learning.