Skip to content

"response.audio_transcript.done" server event arrives too late #25

@fedorenchik

Description

@fedorenchik

Our use case is robot with voice replies and relevant actions. Voice replies are sent via audio channel and relevant robot actions are sent via data channel with "response.create" and "response.done". And "response.create" sends data from "response.audio_transcript.done" server event.

Actual results:

  1. "response.audio_transcript.done" server event arrives 5-35 seconds after relevant audio data finished transmitting (that is, after AI agent stops speaking reply).
  2. "response.done" arrives 5-10 seconds after "response.create" is sent.

So, at the end, robots actions are executed after some delay after robot stops speaking, which looks unnatural and gives an impression that robot is stuck.

Expected results:

  1. "response.audio_transcript.done" server event arrives at the same time or before audio data transmission is finished (robot stops talking)
  2. "response.done" server event arrives somewhat faster? 1-2 seconds if possible.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions