> ## Documentation Index > Fetch the complete documentation index at: https://docs.callers.ai/llms.txt > Use this file to discover all available pages before exploring further. # Conversation dynamics > Control how long your agent waits before replying, when contacts can interrupt, what happens on silence, and how emotional the voice sounds. A good phone conversation depends on timing. These settings decide when the agent starts talking, when it stops for the contact, and how expressive it sounds. Tune them to make the agent feel quick in casual exchanges and patient when the contact needs time to think. Agent-wide settings live in the agent editor under **Agent** (header button); see [Agent settings](https://docs.callers.ai/documentation/build-your-agent/agent-setup.md#agent-voice-and-conversation). Step-level overrides live in each step's settings (the sliders icon on the step).
    The turn-taking loop that these settings control. ## Settings reference | Setting | Where | What it does | Default | | - | - | - | - | | **Patience Level** | **Agent**, and per step | How long the contact must be silent before the agent replies. 100–2000 ms. | 1000 ms | | **Initiate Conversation After No Response** | **Agent** → **Patience Level** | Seconds to wait for the contact before the agent speaks again. Minimum 1 second. | On, 8 seconds | | **Auto-Adjust Patience on Interruptions** | **Agent** → **Patience Level** | Lengthens the wait automatically if the contact keeps getting cut off. | On | | **Speak with no interruptions** | Per step | The agent finishes the step even if the contact talks. | Off | | **Only on first play** | Per step, under **Speak with no interruptions** | Blocks interruptions only the first time the step plays. | Off | | **Voice Emotions** | **Agent** | Lets emotion tags change how the voice sounds. | On automatically for emotion-ready voices | | **Emotion Merge** | **Agent** → **Voice Emotions** | Merges consecutive sentences with different emotions for smoother transitions. | Off | | **Max Merge Sentences** | **Agent** → **Emotion Merge** | Most sentences merged together. 2–5. | 2 | | **Break Time** | **Agent** → **Voice Emotions** | Minimum pause between sentences. 50–2000 ms. | 500 ms | ## Patience level **Patience Level** is how long the agent waits after the contact stops speaking before it replies. The slider runs from **Fast** (about 0.1 s) through **Balanced** (about 1.0 s) to **Slow** (about 2.0 s). * **Faster** settings make small talk and quick yes/no exchanges feel snappy, but the agent may jump in while the contact is still thinking. * **Slower** settings suit questions where the contact needs time: spelling an email address, reading an account number, or checking a calendar. ### Set patience for one step The agent-wide value is the default for every step. To override it: **1. Open the step settings** In the script editor, click the sliders icon on the step. **2. Adjust Patience Level** Expand **Patience Level** and move the slider. Use **Reset to default** to return to the agent-wide value. > [!TIP] > > Keep the agent-wide level near **Balanced**, and slow it down only on steps that ask for details, such as "What's the best email to reach you?" This keeps most of the call fast without cutting people off when they need to think. ### Auto-adjust patience With **Auto-Adjust Patience on Interruptions** on, the agent notices when the contact keeps talking over it and waits longer before replying. Leave it on unless you need fixed, predictable timing. ## When the contact doesn't respond **Initiate Conversation After No Response** sets how many seconds the agent waits for the contact before speaking again, for example to repeat the question or check that the contact is still there. The default is 8 seconds. * Shorten it (4–6 seconds) for quick confirmation calls. * Lengthen it (10–15 seconds) when the contact may need to look something up. If the line stays silent for longer, the agent ends the call. See [Call recording and limits](https://docs.callers.ai/documentation/voice/call-recording-and-limits.md#hang-up-on-silence). ## Interruptions By default contacts can interrupt the agent at any time, just like a person. When the contact starts speaking, the agent stops and listens. ### Uninterruptible steps Some lines must be heard in full, for example a legal disclosure, a recording notice, or a price confirmation. Turn on **Speak with no interruptions** for those steps: **1. Open the step settings** Click the sliders icon on the step. **2. Turn on Speak with no interruptions** The agent will always speak the step fully, even if the contact talks. **3. Optional: allow interruptions on repeats** Turn on **Only on first play** to block interruptions only the first time the step plays. If the agent returns to the step later, the contact can interrupt. For action steps the toggle is called **Speak action response with no interruptions**, and applies to what the agent says with the action's result. > [!NOTE] > > By default, anything the contact says while an uninterruptible step plays is discarded, so the agent doesn't reply to speech it talked over. Contact support if you need to change this. > [!WARNING] > > Use uninterruptible steps sparingly. Long blocks the contact can't interrupt feel robotic and raise hang-up rates. Keep these steps short. ## Emotions and emotion tags Emotion-ready voices can change their delivery line by line: warm for a greeting, apologetic when something went wrong, excited about an offer. In the voice list and the voice library, these voices are marked **Emotions**. When you select an emotion-ready voice, **Voice Emotions** turns on automatically. Turn it off to keep the voice neutral. ### Add emotion tags to your script Add an emotion tag in square brackets before the text it should affect: ```text theme={"dark"} [sympathetic] I'm sorry to hear about the delay. [reassuring] I'll get this sorted for you today. ``` In the script editor, open the tag options under the step text and choose the **Emotions** tab to pick from the list, or type the tag directly. Emotions are grouped by mood, for example **Positive** (happy, warm, encouraging), **Engaged** (curious, sympathetic, concerned), **Uneasy** (hesitant, apologetic), and **Low energy** (calm). The exact list depends on the voice. Tags are not spoken; they only change the delivery. ### Smooth out emotional speech | Setting | Use it when | | - | - | | **Emotion Merge** | Short sentences with different tags sound choppy. Merging produces smoother transitions, at the cost of slightly less precise emotion per sentence. | | **Max Merge Sentences** | You use Emotion Merge and want to limit how many sentences are combined (2–5). | | **Break Time** | Sentences run together or pause too long. Raise it for a slower, more deliberate pace; lower it for a quicker one. | > [!TIP] > > Use one or two emotions per step. Tagging every sentence with a different emotion sounds theatrical. Neutral delivery with a single warm or empathetic tag at the right moment is usually more convincing. ## Tips for natural conversations * **Write short turns.** One idea and at most one question per step. Long monologues invite interruptions and make the agent sound scripted. * **Ask one question at a time.** If you need several details, use separate steps so each has its own patience setting. * **Match patience to the question.** Fast for yes/no, slow for numbers, spelling, and anything the contact has to look up. * **Front-load the important part.** Put the key information at the start of a sentence, so it's heard even if the contact interrupts. * **Keep uninterruptible steps short.** Reserve them for disclosures that must be heard in full. * **Reduce delays from actions.** Run slow in-call actions in the background, or say a short phrase while waiting, so the line isn't silent. See [Actions](https://docs.callers.ai/documentation/build-your-agent/agent-actions.md). * **Test on a real phone.** Timing feels different on a phone call than in a browser. Make several [test calls](https://docs.callers.ai/documentation/build-your-agent/testing-your-agent.md) and use [Insights mode](https://docs.callers.ai/documentation/using-callers/agent-insights.md) to find steps with high interruption rates. ## Troubleshooting > **The agent cuts the contact off mid-sentence** > > Increase **Patience Level** on the step where it happens, or agent-wide. Check that **Auto-Adjust Patience on Interruptions** is on. > **There are long pauses before the agent replies** > > Lower **Patience Level**. If the pause happens after the agent calls an action, the delay may come from the action's response time; run it in the background or add a waiting phrase. > **The agent keeps stopping when there's background noise** > > Turn on or raise [Noise Cancelation](https://docs.callers.ai/documentation/voice/voices-and-sounds.md#noise-cancelation). For a short critical line, use **Speak with no interruptions**. > **The agent repeats itself while the contact is still thinking** > > Increase **Initiate Conversation After No Response** so the agent waits longer before re-engaging. > **Emotion tags have no effect** > > Check that the agent's voice is marked **Emotions** and that **Voice Emotions** is on. Tags must use a supported emotion name in square brackets. ## Related > **[Voices and sounds](https://docs.callers.ai/documentation/voice/voices-and-sounds.md)** > > Choose and tune the agent's voice. > **[Steps](https://docs.callers.ai/documentation/build-your-agent/customizing-steps.md)** > > Build the steps these settings apply to. > **[Keypad input](https://docs.callers.ai/documentation/voice/keypad-input.md)** > > Let contacts answer with the phone keypad. > **[Call recording and limits](https://docs.callers.ai/documentation/voice/call-recording-and-limits.md)** > > Max duration and hang-up on silence. > **[Latency and conversation quality](https://docs.callers.ai/documentation/best-practices/latency-and-conversation-quality.md)** > > Tune patience and handle slow actions without dead air. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.