Microphone capturing the end of the latest response

[wangwillian0]

DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_TTS_START, data={'text': '... simplesmente ficar à disposição para qualquer outra coisa que você precise. O que você gostaria de fazer agora?'}
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_TTS_END, data={'url': 'http://172.18.0.2:8123/api/tts_proxy/q02hd7IuPDJavjL-rSCI2A.mp3'}
DEBUG:linux_voice_assistant.satellite:Playing TTS response: http://172.18.0.2:8123/api/tts_proxy/q02hd7IuPDJavjL-rSCI2A.mp3
INFO:MpvMediaPlayer:Playing 1 URL(s): http://172.18.0.2:8123/api/tts_proxy/q02hd7IuPDJavjL-rSCI2A.mp3
DEBUG:LibMpvPlayer:play: current_state=PlayerState.IDLE
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_RUN_END, data={}
DEBUG:LibMpvPlayer:unduck() called
DEBUG:LibMpvPlayer:_on_start_file: state=PlayerState.LOADING
DEBUG:LibMpvPlayer:_on_end_file: reason=0 (is_eof=True), state=PlayerState.PLAYING, has_callback=True
DEBUG:LibMpvPlayer:_on_end_file: invoking callback
DEBUG:MpvMediaPlayer:Playlist finished, invoking done_callback
DEBUG:linux_voice_assistant.satellite:Continuing conversation
DEBUG:linux_voice_assistant.satellite:TTS response finished
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_RUN_START, data={'url': 'http://172.18.0.2:8123/api/tts_proxy/kdfpVIyoN7QgzvJNVsNcAA.mp3'}
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_STT_START, data={}
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_STT_VAD_START, data={}
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_STT_VAD_END, data={}
DEBUG:linux_voice_assistant.satellite:Voice event: type=VOICE_ASSISTANT_STT_END, data={'text': ' Gostaria de fazer agora.'}

It seems to be happening only on the latest version.

[moritzchow]

I'm also facing the same issue using the latest version.

[florian-asche]

I dont understand. Can you please provide more information?

[moritzchow]

I dont understand. Can you please provide more information?

In short, the STT captured the last 1-2 words of TTS as the Assistant input (it thought its user has spoken)

Simplified Example:
User: "Hello."
STT: "Hello."
Assistant: "Hello, what can I help you today?"
TTS: "Hello, what can I help you today?"
STT: "Today."
Assistant: "Today is Saturday 16th May, 2026. What would you like me to do?"
TTS: "Today is Saturday 16th May, 2026. What would you like me to do?"
STT: "Todo."
Assistant: "..."

So like above the Assistant's STT keeps capturing the TTS's tail 1-2 words, thought it is user who spoke it and then keep continuing the conversation even if the user has no input.

This behaviour only happens when:

1. the user has input that triggers the Assistant thoughts it needs to keep the conversation continued.

2. Then the STT starts listening (From what I see in Home Assistant log, the status of the Assist device changed from "Responding" to "Listening" even before the TTS actually finishes) and capturing the TTS's tail

I can reproduce the issue by just requesting the assistant to ask me a follow up question and then speak nothing thereafter and the Assistant will keep capturing the last 1-2 words of its own TTS output and keep going until it thought no conversation should be continued.

Can provide the HA yaml logs of the history if that helps.