TTS starts after entire text is received

[revmag]

Please read this first

• Have you read the docs?Agents SDK docs
• Have you searched for related issues? Others may have had similar requesrs

Question

Describe your question. Provide details if available.

In the traces dashboard, the post request to audio models is made after the whole post request is received, but why is this so?

Checked the code and it is streaming the audio back for the voice pipeline, and it is partially sending that to the audio model ( event triggered through ResponseCreatedEvent ), so why does the post request to audio models are made after the whole text has been received?

[rm-openai]

@dkundel-openai

[dkundel-openai]

I don't have the timestamps for your particular trace but from what I can see the first TTS request still starts before the Responses API request is fully completed.

In general we can't start the first TTS request until we have sufficient amount of tokens. The way we do this by default is by rudimentary splitting the text by sentences and also making sure the resulting text chunks are longer than 20 characters so that the TTS turns something like "Hi!" into a more natural pronunciation by merging it with the next sentence.

One reason why it can take a while to kick the first one off could be if the Agent responds with a long run-on sentence since we can only start work when that first sentence is completed.

You can modify this behavior by passing your own splitter function into the TTSModelSettings if you would want to experiment with that.

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 3 days since being marked as stale.