Advanced Voice Settings
Overview
The Advanced settings allow administrators to fine-tune how a Voice Agent listens, detects speech, handles interruptions, and manages idle conversations.
These settings are intended for environments where the default behavior requires adjustment due to factors such as:
- Noisy call environments
- Telephony provider characteristics
- User interruption preferences
- Conversation responsiveness
- Voice experience optimization
These settings are intended for advanced users. The default values are suitable for most deployments and should only be changed when there is a specific requirement.
Listening & Response
These settings control how the Voice Agent detects speech and decides when to begin or stop speaking.
Base Response Delay
Specifies how long the assistant waits before beginning its response after detecting that the customer has finished speaking.
Increasing the delay can make conversations feel more natural by avoiding responses during very brief pauses.
Lower values make the assistant respond more quickly.
Noise Filtering
Specifies the minimum speech confidence required before incoming audio is treated as human speech.
Higher values make speech detection stricter and help reduce background noise being interpreted as speech.
Lower values make the assistant more sensitive but may result in false detections in noisy environments.
End of Speech Sensitivity
Specifies the speech confidence below which incoming audio is considered silence.
Lower values allow the assistant to tolerate soft speech and short pauses before assuming the customer has finished speaking.
Higher values cause the assistant to detect the end of speech more aggressively, resulting in faster responses.
Words Required to Interrupt
Specifies the minimum number of spoken words required before an interruption is accepted.
Higher values reduce accidental interruptions caused by short utterances.
Lower values allow faster interruption.
Voice Required to Pause
Specifies how long the customer must continuously speak before the assistant stops talking.
Increasing this value helps ignore very brief sounds or accidental speech.
Reducing this value allows the assistant to stop speaking more quickly when interrupted.
Ignore Early Interruptions
Specifies how long the assistant ignores incoming speech immediately after it begins speaking.
This helps prevent:
- Echo from the assistant's own voice
- Microphone feedback
- Very early accidental interruptions
Increasing this value makes interruptions less sensitive at the beginning of each response.
Interruptions
These settings control how customers can interrupt the assistant while it is speaking.
Allow User Interruptions
Determines whether customers are allowed to interrupt the assistant while it is speaking.
When enabled, the assistant continuously listens for interruptions and stops speaking when the interruption conditions are satisfied.
When disabled, the assistant completes its response before listening again.
Interrupt Keywords
Specifies words or phrases that immediately stop the assistant from speaking without triggering a new response.
Typical examples include:
- stop
- wait
- hold on
- hang on
- cancel that
- never mind
These keywords allow callers to quickly stop the assistant before continuing the conversation.
Ignored Filler Words
Specifies acknowledgement words that should be ignored while the assistant is speaking.
These are common conversational responses that should not interrupt the assistant.
Typical examples include:
- okay
- yeah
- uh huh
- mm-hmm
- got it
- alright
- I see
Ignoring these words helps conversations feel more natural by preventing unnecessary interruptions.
Inactivity & Session Settings
Configure these settings in the Conversational Agent activity under Advanced > Inactivity & Session Settings.
The settings cover two stages of an inactive conversation:
- Idle Detection prompts the user to respond in Web Chat voice sessions and telephone calls. It includes Idle Message, Wait before idle prompt, and Maximum idle prompts. Text-only Web Chat does not receive idle prompts.
- Automatic Conversation End controls when an inactive conversation ends and how long before the end the user is warned. In voice conversations, this period begins after the last idle prompt. In text-only Web Chat, it runs without idle prompts.
For field descriptions, user interaction behavior, an idle message example using Agent Data or Workflow Data, and a timing example, see Inactivity & Session Settings.
Best Practices
The default configuration has been optimized for typical Voice Agent deployments.
Only consider adjusting these settings when:
- customers frequently interrupt unintentionally
- background noise causes false speech detection
- responses feel too early or too delayed
- conversations are ending prematurely
- idle sessions need different timeout behavior
In general, make small adjustments and validate the user experience before applying larger changes.