Any early reviews on scribe v2/realtime

[keen valley]

we are currently rolling out the new feature to some of our clients agents in production while curating the conversations

and planning on sharing our feedback here with you all!

[vast quarry]

in my italian tests, it was superior identifying names, emails etc stuff that wasn't able to recognize before

but on general words seems worst

will keep testing and let you know!

[sharp sapphire]

Anyone had any downsides to it?

[keen valley]

so far not

I guess only people with relly high volume might experience doensides due to latency

and people who do dynamnic multilingual agents

[vast quarry]

latency is better with this new model 😄 overall it seems better but me and my team are still testing it

[keen valley]

here so far results are positive

thinking of enabling it for every hald of our agents and see the results

👀

[vast quarry]

let me know bro !

[tribal spear]

can someone point me to the tool/function calls docs pls? thanks.

[keen valley]

pretty easy to google, but here you go:

• for websockets: https://elevenlabs.io/docs/agents-platform/api-reference/agents-platform/websocket
• general doc: https://elevenlabs.io/docs/agents-platform/customization/tools

[tribal spear]

Thanks.. for some reason I figured realtime v2 may have new docs. Sockets is what I wanted to give it try with. Cheers.

[feral silo]

We have noticed that the model performs worse on common German words, while it handles email addresses and physical addresses much more accurately. For some clients we reverted to the previous v1 setup because the model produced weird unexpected errors.

[keen valley]

that's interesting