Hi, I'm integrating AWS Transcribe with LiveKit AgentSession, and I found that STT only works when allow_interruptions=True.

Here is the setup:

session = AgentSession(
    stt=aws.STT(language="vi-VN"),
    llm=llm_config,
    tts=tts_config,
    vad=ctx.proc.userdata["vad"],
    allow_interruptions=False,
    min_endpointing_delay=0.5,
    turn_detection="vad",
)

When allow_interruptions=True, the transcript is returned correctly.
But when I set allow_interruptions=False, the audio is detected (User is talking (state=speaking)), but no transcript is returned.

This makes it impossible to avoid interrupting the assistant while still getting user input.

Is there any known limitation with allow_interruptions=False and AWS STT?

Does LiveKit drop audio input during the assistant's speaking time or delay it indefinitely?

How can I allow the assistant to finish speaking but still capture what the user said afterward?

Thanks in advance.