> Source: https://builder.ema.ai/v2/builder-guides/voice-ai-employee
> Title: Creating a Voice AI Employee

# Creating a Voice AI Employee

A Voice AI Employee handles phone calls. Callers reach a phone number (or a SIP endpoint), and the AI Employee speaks with them in real time — answering questions, taking actions through your workflow, and transferring to a person when needed. This guide covers provisioning a number, choosing a voice, configuring how the AI Employee talks and behaves, and testing it.

> [INFO]
> **Beta.** Voice AI Employees are in beta. Availability is rolling out gradually; if you don't see voice options when creating an AI Employee, it isn't enabled for your tenant yet — contact your administrator.

## How a Voice AI Employee is structured

A Voice AI Employee is a normal AI Employee with its **interaction type** set to **voice**. Its workflow does the work — intents, knowledge search, tools, transfers — exactly like any other AI Employee. What's special is a companion **voice configuration** that bridges telephony to that workflow:

-   Telephony (a phone number via Twilio, or a SIP endpoint) connects the caller.
-   A real-time voice platform handles speech-to-text and text-to-speech and runs the conversation.
-   During the call, the AI Employee dispatches tools and can transfer the caller to a person.

When you create a voice AI Employee, the companion voice configuration is provisioned automatically. You then tune it on the AI Employee's **Voice settings** tab.

## Step 1 — create the Voice AI Employee

Create a new AI Employee and choose the voice modality (the voice starter template). This sets the interaction type to `voice` and provisions the voice configuration alongside the workflow. Build the workflow itself the same way as any other AI Employee — see [Build your first workflow](/builder/v2/builder-guides/build-your-first-workflow).

> [TIP]
> If the voice configuration ever fails to provision (for example, the voice platform was briefly unavailable when the AI Employee was created), the Voice settings tab shows a banner with a **Retry** action that re-runs provisioning.

## Step 2 — provision a phone number (or SIP endpoint)

Open the Voice settings tab. The **Phone numbers** card is where callers reach the AI Employee. You have two options, and they're mutually exclusive per AI Employee:

-   **Provisioned number** — search for and provision a phone number through the built-in provider, then assign it to this AI Employee. You can give each number a friendly display name.
-   **SIP endpoint** — connect your own telephony via SIP, in one of two modes:
    -   **IP allowlist** — accept SIP traffic from specified CIDR ranges.
    -   **Registration** — register with a SIP username, password, and proxy.

Add a number with the **\+ Add** action; remove one from its row menu. Switching to SIP is offered when no provisioned number is set, and vice versa.

## Step 3 — choose a voice and language

In the **Voice** card, set:

-   **Language** — the primary language for the call, from the available catalog.
-   **Voice** — pick from the voices available for that language. Each voice card shows a description, the languages it supports, and a preview you can play.
-   **Model** — the underlying voice model (defaults to a sensible value; change only if you have a reason to).

## Step 4 — configure the conversation

The **Conversation** card defines how the AI Employee speaks:

-   **System prompt** — the voice role and framing (who the AI Employee is and how it should sound).
-   **Welcome message** — the greeting played when a call connects. Insert the `{{voice_name}}` variable to drop in the AI Employee's name (for example, "Hello, this is {{voice\_name}}, how can I help you?").
-   **Hangup message** — what's said before the call ends.

> [TIP]
> Keep voice copy short and spoken-natural. Long sentences that read fine on screen sound stilted when spoken. Write the welcome and hangup messages the way a person would actually say them on the phone.

## Step 5 — set call and detection behavior

Several cards tune how the AI Employee handles the mechanics of a call:

-   **Tools** — descriptions for the voice platform's built-in tools, including whether the AI Employee can disconnect the call (and the instructions for doing so), a knowledge corpus id for fast lookups, and instructions for the fast-query and take-action tools.
-   **Call settings** — toggles for call forwarding and spam prevention.
-   **Voice activity detection (VAD)** — the turn timeout and silence timeout (in milliseconds) that govern when the AI Employee decides the caller has finished speaking.
-   **Data storage** — whether to store the call audio recording, the caller transcript, and the agent transcript.

Each card has its own **Edit** / **Save** controls, so you can change one area at a time.

## Step 6 — route calls to people

The **Call forwarding destinations** card lists the people or teams the AI Employee can transfer a caller to. For each destination, add:

-   **Label** — a name for the destination (for example, "Billing team").
-   **Phone number** — where to transfer.
-   **Description** — optional context the AI Employee uses to decide when this destination applies.
-   **Transfer type** — **cold** (hand off and drop) or **warm** (announce the caller before connecting).
-   **Availability** — optionally a timezone and weekly schedule windows, so the destination is only offered when it's staffed. With no schedule, the destination is always available.

The AI Employee picks the right destination during the call based on the destination labels and descriptions, then transfers using the voice platform's built-in transfer tools. Adding the first destination enables call forwarding; removing the last one disables it.

## Step 7 — test the call

The **Preview call** card lets you talk to the AI Employee from your browser — no phone needed. Start the preview, allow microphone access, and have a conversation; you can mute and unmute your mic, and see the live transcript. Use this to validate the voice, greeting, and conversational flow before pointing real callers at it.

## After the call

Completed calls are recorded in the AI Employee's **Voice audit** view, with the caller and called numbers, start time, duration, a summary, who ended the call, and — when data storage is enabled — the recording and full transcript. Use it to review how calls went and spot where the AI Employee needs better Instructions or an additional transfer destination.

## What's next

-   [Build your first workflow](/builder/v2/builder-guides/build-your-first-workflow) — the workflow behind the voice.
-   [Writing effective Instructions](/builder/v2/builder-guides/writing-instructions) — guide the agents your voice AI Employee runs.
-   [Designing human-in-the-loop forms](/builder/v2/builder-guides/designing-hitl-forms) — voice uses conversation-mode HITL to collect input from callers.