8. Agent Configurations

• TTS Caching Strategy: decides how the agent’s speech is saved (cached) during a call to improve speed and audio performance. Full Response Caching saves the whole response (lowest latency, best for repeated scripted lines). No Caching builds audio fresh every time (most flexible, but higher latency). Sentence-Level Caching saves individual sentences (a balance of low latency and dynamic responses).
• Default Language: the main language the agent uses during the call.
• Silence Message Count: how many times the agent repeats the silence prompt before ending the call.
• Wait Till User Reply (sec): how long the agent waits for a reply before triggering a silence prompt.
• Silence Message Language: the language used for silence prompts and follow-up messages.
• Silence Mode: controls how the agent handles silent periods during a conversation.
• Detect Gender (ON/OFF): turns gender detection on or off to personalise responses based on the caller’s voice.
• Silence Duration (ms): how much silence is needed before the agent decides the user has stopped speaking.
• Threshold: sets how sensitive voice detection is, to tell speech apart from background noise.
• Prefix Padding (ms): adds a small audio buffer before detected speech, so the start of words is not cut off.
• Minimum Speech Duration (sec): the shortest speech length that counts as valid user input.
• Interrupt Speech Duration (sec): how long a user must speak to interrupt the agent’s response.
• Interrupt Min Words: the minimum number of words needed to interrupt the agent while it is speaking.
• Clear Buffer if Not Interrupted (ON/OFF): turns on or off the automatic clearing of buffered speech when no interruption happens.
• Manual Interruption (ON/OFF): turns manual interruption control on or off.
• Manual Interruption Speech Duration (sec): the speech length needed to trigger a manual interruption.
• Max Buffered Speech (sec): the most user speech that can be stored in the buffer.
• Min Endpointing Delay (sec): the minimum wait before the agent decides the user has finished speaking.
• Max Endpointing Delay (sec): the maximum the agent waits before ending the user’s speaking turn.
• Preemptive Synthesis (ON/OFF): turns early response generation on or off to reduce felt latency.
• Language Switching (ON/OFF): turns automatic language switching on or off, based on the user’s recent messages.
• Language Switching V1 (ON/OFF): switches language only after getting the user’s confirmation or consent.

​

Monitor the configured agent