> ## 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.

# Latency and conversation quality

> Make voice agents feel responsive: tune patience, handle slow actions, reduce noise and interruptions, and diagnose issues from transcripts and insights.

A voice agent feels natural when it answers promptly, doesn't talk over the contact, and doesn't leave long silences. This page explains what affects responsiveness on Callers, which settings you control, and how to find the cause when a call feels slow or choppy.

## What affects responsiveness

Every agent turn has three parts: detecting that the contact has finished speaking, generating a response, and turning it into speech. Several things you configure add time to that loop.

| Factor | Effect | What you can do |
| - | - | - |
| **Patience Level** | How long the agent waits after the contact stops talking before it replies. Higher values add a deliberate pause to every turn. | Lower it for quick conversations; raise it per step where contacts need time to think. |
| **In-call actions** | A step that calls an API or App Center action waits for the response before the agent continues. | Run slow actions in the background or use a **Wait** step with a message. Keep APIs fast. |
| **Knowledge base lookups** | Retrieval adds time when the agent searches for an answer. Large, unfocused content makes answers slower and less precise. | Keep knowledge bases small and topical. Put frequent answers in scenarios. |
| **Long steps and instructions** | Long prompts and long spoken turns take longer to generate and to speak, and invite interruptions. | Keep steps to one to three sentences. Keep Agent Instructions focused. |
| **Telephony route** | The network path between the carrier and Callers adds delay, especially for SIP trunks far from where calls are processed. | If you bring your own carrier over SIP and calls feel slow, contact [support@callers.ai](mailto:support@callers.ai) to review routing. |
| **Greeting start** | On outbound calls, the first words can lag slightly behind the contact answering. | Keep the greeting short. Contact support to adjust greeting preloading for your agent. |

## Tune patience

**Patience Level** controls how long the agent waits after the contact stops talking. Open the agent's voice settings and expand **Patience Level**.

| Setting | What it does | Default |
| - | - | - |
| **Patience Level** | Slider from **Fast** (about 0.1 s) through **Balanced** (about 1 s) to **Slow** (about 2 s). Range 100–2000 ms. | 1000 ms |
| **Initiate Conversation After No Response** | Seconds of silence before the agent speaks again, for example to check the contact is still there. Minimum 1 s. | 8 s |
| **Auto-Adjust Patience on Interruptions** | The agent adjusts its delay if interruptions happen too often. | On |

You can also override patience on a single step from the step's settings. A common pattern:

* **Fast** for greetings, confirmations, and small talk.
* **Balanced** for most of the script.
* **Slow** for steps where the contact reads out a number, spells a name, or thinks about a decision.

> [!TIP]
>
> If contacts often get cut off mid-sentence, raise patience. If the agent feels sluggish after short answers like "yes," lower it. Change it in small steps and test after each change.

## Handle slow actions without dead air

When an in-call action takes more than a moment, the contact hears silence. Use one of these patterns.

**Run in background**

In the action step menu, turn on **Run in background without pausing the conversation**. The agent keeps talking and the response can arrive later in the conversation.

Use it when the conversation doesn't depend on the result right away, for example logging to a CRM or starting a lookup the agent will need a few turns later.

> [!NOTE]
>
> If you don't see this option in the action step menu, contact [support@callers.ai](mailto:support@callers.ai) to turn it on.

**Wait step**

Add a **Wait** step after the action. Configure:

* **Say while waiting**: what the agent says while the action runs, such as "One moment while I check that."
* **Wait for**: the actions (or connected agents) to wait on.
* **Continue when**: all or any of them finish.
* Timeout: 5–180 seconds, default 60.
* **Paths**: a ready path when results arrive, and a fallback path on timeout.

Use it when the next step depends on the result.

**Action phrase**

Give the action a short phrase the agent says as it starts the action, such as "Let me pull that up." This covers a short delay without a separate step.

Turn on **Allow action without phrase** only for actions that return almost instantly.

Also make the API itself fast: return only the fields the agent needs, avoid chaining several slow requests in one action, and add a fallback phrase for errors and timeouts. See [Actions](https://docs.callers.ai/documentation/build-your-agent/agent-actions.md).

## Interruptions

Contacts interrupt for good reasons: they already know the answer, they disagree, or the turn is too long. Some interruptions are accidental, such as background voices or "mm-hm."

* **Shorten turns first.** Long turns are the most common cause of interruptions.
* **Protect only what must be heard.** Turn on **Speak with no interruptions** for disclosures and required lines. Add **Only on first play** if the step can repeat.
* **Use Auto-Adjust Patience on Interruptions** so the agent adapts when a contact interrupts often.
* **Check the step in Insights mode.** A high interruption rate on one step usually means it is too long or asks too much.

See [Conversation dynamics](https://docs.callers.ai/documentation/voice/conversation-dynamics.md).

## Background noise

Noise from traffic, TV, or other people can trigger false interruptions or make speech harder to recognize.

* **Noise Cancelation** (agent voice settings): reduces unwanted sounds such as chatter and keystrokes. Choose a level from **Basic** to **Extreme**. Higher levels remove more noise but can affect quiet speech, so test with realistic calls.
* **Keypad input** is more reliable than speech for digits in noisy settings.
* **Speech recognition vocabulary** helps the agent recognize product names and jargon. See [Speech recognition vocabulary](https://docs.callers.ai/documentation/voice/speech-recognition-vocabulary.md).

> [!NOTE]
>
> If you don't see **Noise Cancelation** in your agent's voice settings, contact [support@callers.ai](mailto:support@callers.ai) to turn it on.

> [!WARNING]
>
> Background soundscapes (for example, office ambience) are only available on SIP calls. They make the agent sound more natural but add audio the contact hears. Keep the volume low.

## Diagnose a slow or choppy call

**1. Listen to the recording**

Open the call from the agent's results. Listen for long gaps, the agent talking over the contact, or the agent cutting the contact off. Recordings show timing that transcripts can't.

**2. Read the transcript with timing details**

In the transcript, hover the info icon next to a message to see timing details. For agent messages, this shows how long the response took to start generating and to start speaking. For contact messages, it shows speech recognition confidence. Low confidence points to noise, a poor line, or words to add to the vocabulary.

**3. Look for tool calls**

Find action calls in the transcript. A long gap right after an action means the API is slow. Move it to the background, add a Wait step, or speed up the API.

**4. Check Insights mode**

Turn on **Insights Mode** in the agent header to see, per step, how many calls reached it, where contacts hang up, and where interruptions are frequent. Fix the steps with the most drop-offs and interruptions first.

**5. Report the issue**

If a call shows a problem you can't fix with settings, report it from the call view. Choose a category such as **Speech Interruptions** or **Speech Recognition**. See [Reporting call issues](https://docs.callers.ai/documentation/using-callers/reporting-call-issues.md).

## Troubleshooting

> **The agent waits too long before answering**
>
> Lower **Patience Level**, or lower it on the specific steps where answers are short. Check whether the step runs an action; if so, run it in the background or add a Wait step with a message.

> **The agent cuts the contact off**
>
> Raise **Patience Level** for that step, especially where contacts give long or thoughtful answers. Make sure **Auto-Adjust Patience on Interruptions** is on.

> **The agent keeps getting interrupted**
>
> Shorten the step. If the line must be heard, turn on **Speak with no interruptions**. If the call is noisy, try **Noise Cancelation**.

> **The agent says 'are you still there?' too soon**
>
> Increase **Initiate Conversation After No Response**. Make sure the previous step ends with a clear question so the contact knows it is their turn.

> **Answers from the knowledge base are slow or wrong**
>
> Split large documents into smaller, topical items with clear headings. Remove duplicate content. For questions that come up on most calls, write a scenario.

> **The agent mishears names or terms**
>
> Add them to the [speech recognition vocabulary](https://docs.callers.ai/documentation/voice/speech-recognition-vocabulary.md). For digits, use keypad input.

## Related

> **[Conversation dynamics](https://docs.callers.ai/documentation/voice/conversation-dynamics.md)**
>
> Patience, interruptions, and no-response behavior.

> **[Prompting for voice](https://docs.callers.ai/documentation/best-practices/prompting-voice-agents.md)**
>
> Write steps that are quick to say and easy to follow.

> **[Actions](https://docs.callers.ai/documentation/build-your-agent/agent-actions.md)**
>
> In-call actions, background runs, and Wait steps.

> **[Agent results and insights](https://docs.callers.ai/documentation/using-callers/agent-insights.md)**
>
> Find drop-offs and interruptions by step.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.