wx44wx/pipecat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Ignore Gemini VAD interrupts when VAD is disabled

Browse Source 
When GeminiVADParams(disabled=True) is set, Pipecat correctly sends the
disabled flag to Google's API. However, the API may still send interrupt
signals despite this setting. Without a local guard, these interrupts
are unconditionally broadcast, causing the bot to interrupt users even
when the developer explicitly disabled server-side VAD.

This adds a check to suppress interrupt broadcasting when VAD is
configured as disabled, logging the ignored signal at debug level.
This commit is contained in:
James Hush
2026-03-25 13:47:17 +08:00
committed by Mark Backman
parent b80328e038
commit 13fcf966a4
1 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/pipecat/services/google/gemini_live/llm.py
View File

@@ -828,7 +828,6 @@ class GeminiLiveLLMService(LLMService):
if self._settings.language if self._settings.language
else "en-US" else "en-US"
) )
self._vad_params = self._settings.vad
# Reconnection tracking # Reconnection tracking
self._consecutive_failures = 0 self._consecutive_failures = 0
@@ -1371,8 +1370,13 @@ class GeminiLiveLLMService(LLMService):
# track turns, such as local Silero VAD in # track turns, such as local Silero VAD in
# combination with the context aggregator default # combination with the context aggregator default
# turn strategies. # turn strategies.
logger.debug("Gemini VAD: interrupted signal received") if self._settings.vad and self._settings.vad.disabled:
await self.broadcast_interruption() logger.debug(
"Gemini VAD: interrupted signal received (ignored, VAD disabled)"
)
else:
logger.debug("Gemini VAD: interrupted signal received")
await self.broadcast_interruption()
elif message.server_content and message.server_content.model_turn: elif message.server_content and message.server_content.model_turn:
await self._handle_msg_model_turn(message) await self._handle_msg_model_turn(message)
elif message.server_content and message.server_content.turn_complete: elif message.server_content and message.server_content.turn_complete: