Order between UserStoppedSpeakingFrame from VAD and TranscriptionFrame from TTS

[zizhong]

Description

Previously our logic used to assume UserStoppedSpeakingFrame comes after TranscriptionFrame. So if user stopped speaking, we will get all the text from TranscriptionFrames and use UserStoppedSpeakingFrame as a turn indicator.

However, I found now it is different. UserStoppedSpeakingFrame can come before TranscriptionFrame.
Is it a bug or expected behavior? If expected, what is the recommanded turn indicator?

If reporting a bug, please fill out the following:

Environment

• pipecat-ai version: 0.0.50
• python version: 3.10
• OS: ubuntu

Issue description

UserStoppedSpeakingFrame can come before TranscriptionFrame.

Repro steps

Pipeline with

• FastAPIWebsocketTransport + SileroVADAnalyzer
• STT service

Expected behavior

UserStoppedSpeakingFrame as a turn indicator.

Actual behavior

UserStoppedSpeakingFrame is not a turn indicator.

Logs