[(1) ]() Lower both values from their defaults toward 0 to make the agent wait longer before speaking and harder to interrupt by ambient noise, giving the user more room to finish speaking. Separately, note that Retell's turn-taking model is designed to wait if it detects the user hasn't finished talking, based on contextual cues rather than just pauses [(4) ]() [(5) ]() . If interruptions persist, "High Overlapping Speech Count" issues can also stem from latency — if end-to-end latency (p50) exceeds 2.5s, fixing latency first is recommended, since users are more likely to talk over the agent when it's slow to respond [(6) ]() .

### Answer To reduce agent interruptions, adjust two agent settings: - **Decrease `responsiveness`**: this controls how quickly the agent responds — lower values mean the agent waits more and responds slower [(1) ]() [(2) ]() - **Decrease `interruption_sensitivity`**: lower values mean it takes longer/more words for the user to interrupt the agent; setting it to 0 means the agent would never be interrupted For a case where an agent was interrupting users mid-sentence (especially in the information-gathering phase) , the recommended fix was to decrease Responsiveness and increase Interruption Sensitivity — note this is the opposite direction depending on which side has the problem (agent interrupting user vs. user interrupting agent hard to trigger) [(3) ]() . In your case, since the agent is interrupting the user, you want to decrease

responsiveness

and decrease

interruption_sensitivity

[(1) ]() [(2) ]() . You can update these via the API:

json
"responsiveness": 1,
"interruption_sensitivity": 1,

​ To learn more, see the sources I considered: [[1] Get Voice Agent](https://docs.retellai.com/api-references/get-agent) [[2] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[3] Agent interrupting users on speakerphone calls](https://discord.com/channels/1186424376300679209/1442882951246512168) [[4] VAD vs. Turn-Taking Endpoints in Conversational AI | Retell AI](https://www.retellai.com/blog/vad-vs-turn-taking-end-point-in-conversational-ai) [[5] Retell AI's Turn-Taking Model for Seamless Calls | Retell AI](https://www.retellai.com/blog/how-retell-ais-turn-taking-model-ensures-seamless-calls) [[6] Address metric issues](https://docs.retellai.com/ai-qa/address-metric-issues) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Retell AI's turn-taking model doesn't just detect speech—it determines when to respond and when to wait based on context and intent, preventing interruptions by recognizing subtle cues like tone shifts, pauses, and sentence patterns [(1) ]() . Key features: - **Contextual Analysis**: Combines sound signals with semantic understanding to determine whether a user is pausing or finished speaking - **No Interruptions**: Waits patiently if the user hasn't finished talking, reducing the risk of cutting them off mid-sentence - **Adaptive Responses**: Learns from diverse datasets to handle different speaking styles and environments, even in noisy or multi-speaker settings - **Natural Flow**: Maintains conversation continuity, making interactions feel human-like and effortless Unlike Voice Activity Detection (VAD) alone, which only identifies speech boundaries and can lead to interruptions if pauses are misinterpreted as the end of a user's turn, Retell AI's turn-taking model prevents interruptions by recognizing subtle cues like tone shifts, pauses, and sentence patterns [(2) ]() . More broadly, Turn-Taking Endpoints are the mechanisms that determine when a speaker (human or AI) has finished talking and it's appropriate for the other party to begin speaking [(3) ]() . In AI voice systems, where nonverbal cues are absent, turn-taking must be detected and managed precisely, or conversations break down . Key factors include Voice Activity Detection (detecting when the user is speaking, pausing, or has finished) , Pause Duration Thresholds (how long a silence must last before the AI concludes the user is done speaking) , Speech Patte

rns and Prosody Analysis (recognizing rising intonation versus final statements) , and Interruption Handling (detecting when the user starts speaking while the AI is talking and handing the floor back gracefully) . In practice: a customer says, "I need…uh, wait, one second…yeah, I need to change my delivery address." The voice agent, using VAD and turn-taking endpoints, recognizes the hesitation and only responds after the full request is completed, avoiding cutting off the customer mid-thought .

​ To learn more, see the sources I considered: [[1] VAD vs. Turn-Taking Endpoints in Conversational AI | Retell AI](https://www.retellai.com/blog/vad-vs-turn-taking-end-point-in-conversational-ai) [[2] Why Low Latency Matters for AI Agents | Retell AI](https://www.retellai.com/blog/why-low-latency-matters-how-retell-ai-outpaces-traditional-players) [[3] Turn-Taking Endpoints](https://www.retellai.com/glossary/turn-taking-endpoints) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer The available sources don't explicitly document a dedicated on/off toggle specifically for "the turn-taking model" as its own setting. What they do document is that agents have configurable Responsiveness and Interruption Sensitivity settings, which directly influence turn-taking behavior — how long the agent waits and how easily it can be interrupted [(1) ]() . - **Responsiveness**: controls how responsive the agent is — set lower to make it wait longer before responding, or check "Dynamically adjust based on user input" to let the agent auto-tune timing based on how fast the user speaks . - **Interruption Sensitivity**: controls how easily the agent gets interrupted by the user — set lower for more resilience to background speech . Via the API, these correspond to the

responsiveness

and

interruption_sensitivity

fields:

json
"responsiveness": 1,
"interruption_sensitivity": 1

[(2) ]()

responsiveness

ranges [0,1], where lower means the agent waits more and responds slower, and if unset, defaults to 1 .

interruption_sensitivity

ranges [0,1], where lower means it takes longer/more words for the user to interrupt the agent, and setting it to 0 means the agent would never be interrupted; if unset, defaults to 1 . I don't have documentation confirming a separate, standalone "turn-taking model" toggle beyond these two settings.

​ To learn more, see the sources I considered: [[1] Step 2: Configure basic agent settings](https://docs.retellai.com/build/single-multi-prompt/configure-basic-settings) [[2] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.