Conversation Timing
Turn Detection Eagernessโ
Controls how quickly the AI detects when the user has finished speaking.
Options:
- High (most responsive): The agent responds quickly, which feels snappy but may occasionally interrupt if the customer pauses mid-sentence. This is the default.
- Medium (balanced): A balanced approach that waits a bit longer before responding.
- Low (most patient): The agent waits longer to ensure the customer is truly finished speaking. Use this if customers frequently get interrupted.
When to adjust:
- If customers complain about being interrupted, lower the eagerness
- If conversations feel slow or awkward due to long pauses, increase the eagerness
- For customers who speak slowly or take pauses while thinking, use Low
Default: High
Initial Mute Durationโ
Duration in milliseconds to keep the caller muted at the start of the conversation.
This allows your agent to introduce itself without interruption, which is especially useful for:
- Longer greeting messages
- Agents that provide important upfront information
- Scenarios where you want to ensure the introduction is heard completely
Measure your introduction to find the right duration.
Example values:
0- No mute (customer can speak immediately)5000- 5 seconds (recommended for standard greetings)8000- 8 seconds (for longer introductions)
Default: 0 (disabled)
Silence Timeoutโ
Controls how long the agent waits in silence before attempting to re-engage the caller.
When neither the caller nor the agent has spoken for the configured duration, the agent is prompted to evaluate the situation and decide whether to check in โ for example, asking if the caller is still there, repeating the last question, or offering help.
Default: 9 seconds
When to adjust:
- If callers frequently need time to look something up, increase the timeout to avoid unnecessary interruptions
- If dead air feels awkward or callers often go quiet unexpectedly, lower the timeout so the agent re-engages sooner
The silence timer only runs during active calls. It pauses while the agent is speaking, processing a response, or waiting for a function call to complete.