> For the complete documentation index, see [llms.txt](https://docs.limecall.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.limecall.com/ai-receptionist/voice-and-tone.md).

# Voice & tone

How your receptionist sounds.

Open **AI Receptionist → Voice & Tone**.

## Choosing a voice

Six voices are available:

| Voice      | Character                    |
| ---------- | ---------------------------- |
| **Olivia** | Warm, friendly · female      |
| **Sophie** | Bright, upbeat · female      |
| **Alex**   | Even, neutral · male         |
| **Ethan**  | Calm, measured · male        |
| **Freya**  | Expressive, British · female |
| **Oscar**  | Deep, authoritative · male   |

Listen to each before choosing. A voice that reads well as a description can be wrong for your callers.

Match the voice to the situation rather than to personal taste:

* **Calm and measured** (Ethan, Alex) suits calls where people are stressed — medical, emergency, complaints.
* **Warm and friendly** (Olivia) suits consumer services and hospitality.
* **Bright and upbeat** (Sophie) suits retail and consumer sales, and grates in a serious context.
* **Authoritative** (Oscar) suits professional services.
* **British** (Freya) matters if your customers are British; an American voice on a UK local number gets commented on.

## Response speed

Three settings:

**Fast — reply quickly.** Short, transactional calls. Risks cutting people off.

**Balanced (recommended).** Where to start.

**Patient — wait for pauses.** For callers who speak slowly, pause to think, or read things out. Choose this for older demographics, and for any call where people recite a policy number or an address.

This setting has more effect on how natural the call feels than the voice does. If callers report being interrupted, move toward Patient before changing anything else.

## Tone of voice

Written instructions about how it should speak. These apply on written channels too.

Be specific and give examples:

> Speak plainly. Short sentences. Do not say "absolutely" or "perfect". If you do not know something, say "I'm not sure, but I can find out and have someone call you back" rather than guessing.

Naming the words you do not want is unusually effective — a generic instruction to "be natural" does far less than a short list of banned filler.

## Pronunciation

If the assistant mispronounces a name, a product or a place, add it under **Specialized terms** in [Scenarios & call types](/ai-receptionist/scenarios-and-call-types.md). This helps both output and recognition.

## Testing the voice

Always test on a real phone, not only in the browser. Phone audio is narrowband and compressed, and voices that sound clear on laptop speakers can be harder to follow down a line.