#Orchestration question

1 messages · Page 1 of 1 (latest)

hollow lark
#

Which transcriber are you using, default or Scribe?

#

Can you share some conv ids?

#
  1. there is a silence before initiating the turn as well that is not displayed - do you have speculative turn on?
#
  1. can you share some conversation id
#
  1. we fire the pre tool message as soon as we receive from LLM intent to call a tool, if tool is fast by the time we generate audio for the pre-tool, we are already generating response - this we can optimize a bit actually have the audio be prerendered for tool, btw. try using pre-tool speech auto mode
shut pasture
#

Using Scribe as it was documented have best accuracy. But when trying default get better results.
2) have this turned on
3) conv_3701khb2e95hegzvqnbvd09zhsza

conv_9101khb3ksfye1vbmgd8ej8ecbrh
4) tried but still seems it takes longer to generate pre tool speech than after tool give response