Hey everyone

[restive spear]

We tried the new scribe v2 as well and had the same issue. It detected most words much worse than the scribe v1. Also in our tests it seemed like the keyword boosting does not work with the v2.

So we switched back to v1 with keyword boosts which performs much better and more stable than v2. I am hoping for some kind of update or benchmarks as it seems v2 is not really working as good as in the benchmarks and claims by elevenlabs.

[solemn bolt]

same here unfortunately, was really hyped for it

[rocky karma]

Thanks for the feedback! The thing is I don't know in advance what terms users will say - could be any email address, company name, product name, acronym etc. So I can't really populate keywords with everything possible

Every time I try to demo with someone, they get frustrated by this and stop testing

[solemn bolt]

personally i wouldnt tell you r clients the ai can already transcribe email addresses etc. it's too dynamic atm, you can have a lot of words in an email the transcription model isnt trained on. it will come but not now unfortunately