Thanks — I looked into the BackgroundAudioPlayer behavior you described. Short answer: There is no built-in way to scope

BackgroundAudioPlayer.thinking_sound

to only tool/function calls or to set a

min_duration_ms

.

thinking_sound

fires whenever the agent enters the "thinking" state (which happens for any LLM turn, not just tool execution). Workarounds you can use: • Remove

thinking_sound

from the automatic BackgroundAudioPlayer config and play audio manually inside your tool functions using

background_audio.play()

(or

context.session.say()

if you prefer TTS). This ensures the typing/hold sound only runs during actual tool execution. • Pre-synthesize and cache any audio clips to avoid TTS latency when playing them during tool calls. • If needed, use

context.disallow_interruptions()

while playing a tool-specific audio cue to prevent premature interruption. If you want, tell me which SDK/language you’re using and I’ll give a short code example for playing the audio only inside tool execution. Sources: Background audio | LiveKit Documentation | Turns overview | LiveKit Documentation | External data and RAG | LiveKit Documentation | Function tool definition | LiveKit Documentation | Audio customization | LiveKit Documentation, with interactive elements