Voice Response Settings | Levrage.AI

The voice_response_settings object lets you control every aspect of your agent’s conversational behavior — from turn-taking sensitivity to call duration limits.

Pass it when creating or updating an agent:

curl -X PATCH https://api.levrage.ai/v1/agents/{agent_id} \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "voice_response_settings": {
      "allow_user_to_interrupt": true,
      "words_to_interrupt": 2,
      "silence_before_reminder": 8,
      "max_call_duration_minutes": 60,
      "response_creativity": 0.8
    }
  }'

Turn-Taking & Interruption

Control when the agent speaks and how it handles user interruptions.

Field	Type	Default	Description
`allow_user_to_interrupt`	boolean	`true`	Allow user to interrupt agent mid-speech
`words_to_interrupt`	integer (0–10)	`2`	Minimum words spoken before an interruption is accepted
`turn_detection`	string	`"vad"`	Turn detection mode: `vad` (voice activity), `stt` (speech-to-text), `semantic` (AI-based), `english`
`vad_type`	string	`"server_vad"`	VAD type: `server_vad` (silence-based) or `semantic_vad` (AI-based turn detection)
`eagerness`	string	`"medium"`	For `semantic_vad` mode: `low`, `medium`, `high`, or `auto`

Response Timing

Fine-tune how quickly the agent responds after the user stops speaking.

Field	Type	Default	Description
`wait_before_responding`	integer (50–3000)	`200`	Padding in milliseconds before agent starts speaking
`silence_to_end_turn`	integer (100–3000)	`300`	Silence duration (ms) to detect end of user turn
`min_response_wait`	float (0.05–1.0)	`0.15`	Minimum seconds to wait before responding
`max_response_wait`	float (0.1–2.0)	`0.3`	Maximum seconds to wait before responding
`fast_response`	boolean	`true`	Enable preemptive generation for faster responses
`quick_response_padding`	integer (50–1000)	`200`	Response padding (ms) for realtime models
`turn_end_silence`	integer (100–2000)	`300`	Silence to detect end of turn in realtime models

Listening Sensitivity

Adjust how sensitive the agent is to voice input.

Field	Type	Default	Description
`listening_sensitivity`	float (0.1–1.0)	`0.5`	Voice detection sensitivity. Lower = more sensitive (picks up soft speech). Higher = needs louder/clearer voice
`voice_detection_sensitivity`	float (0.1–0.9)	`0.5`	Voice detection threshold for premium models

Silence & Reminders

Configure what happens when the user goes silent during a call.

Field	Type	Default	Description
`remind_on_silence`	boolean	`true`	Send reminders when user is silent
`silence_before_reminder`	integer (3–60)	`8`	Seconds of silence before sending a reminder
`max_reminders`	integer (1–10)	`5`	Max reminders before auto-disconnect
`custom_reminder_messages`	array[string]	—	Custom messages. E.g. `["Hello, are you there?", "Please say something..."]`

Example

{
  "remind_on_silence": true,
  "silence_before_reminder": 10,
  "max_reminders": 3,
  "custom_reminder_messages": [
    "Hello, are you still there?",
    "I'm still here if you'd like to continue.",
    "Alright, I'll end the call now. Goodbye!"
  ]
}

Call Control

Manage call duration, IVR detection, and handoff behavior.

Field	Type	Default	Description
`max_call_duration_minutes`	integer (1–180)	`60`	Auto-end call after this many minutes
`greeting_delay`	integer (0–10)	`0`	Delay in seconds before agent greets on outbound calls
`detect_ivr_systems`	boolean	`true`	Auto-detect IVR/voicemail systems on outbound calls
`enable_dtmf_input`	boolean	`false`	Enable DTMF keypress detection (e.g. “press 1 for sales”)
`leave_voicemail`	boolean	`false`	Leave a voicemail when an answering machine is detected
`immediate_transfer`	boolean	`false`	Immediately transfer to a human without AI interaction

AI & Quality

Control AI behavior and post-call analysis.

Field	Type	Default	Description
`response_creativity`	float (0.0–1.0)	`0.8`	AI creativity level. `0` = focused/deterministic, `1` = creative/varied
`call_analysis_enabled`	boolean	`true`	Enable post-call AI analysis (sentiment, summary, data extraction)

Recommended Presets

Sales Agent

Fast, interruptible, with analysis enabled.

{
  "allow_user_to_interrupt": true,
  "words_to_interrupt": 1,
  "fast_response": true,
  "response_creativity": 0.9,
  "max_call_duration_minutes": 15,
  "call_analysis_enabled": true
}

Support Agent

Patient, high sensitivity, longer calls.

{
  "allow_user_to_interrupt": true,
  "words_to_interrupt": 3,
  "silence_before_reminder": 12,
  "max_call_duration_minutes": 30,
  "response_creativity": 0.6,
  "listening_sensitivity": 0.3
}

Start with the defaults and adjust based on your use case. The most impactful settings are response_creativity, silence_before_reminder, and max_call_duration_minutes.