> ## Documentation Index
> Fetch the complete documentation index at: https://docs.itellico.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Voice Cloning

> Create custom cloned voices from audio samples or recordings for your AI agents

export const Screenshot = ({lightSrc, darkSrc, alt, caption, maxWidth = "880px"}) => {
  return <div style={{
    margin: "1rem auto",
    maxWidth,
    width: "100%"
  }}>
      <Frame>
        <img className="block dark:hidden" src={lightSrc} alt={alt} />
        <img className="hidden dark:block" src={darkSrc} alt={alt} />
      </Frame>
      {caption ? <p style={{
    marginTop: "0.5rem",
    fontSize: "0.875rem",
    color: "inherit",
    opacity: 0.8
  }}>
          {caption}
        </p> : null}
    </div>;
};

<Badge>Expert Mode</Badge>

Voice cloning lets you create a custom synthetic voice from audio and use it with your AI agents. You can either **upload** a recording or **record** directly in the browser. The platform generates a voice model through your chosen provider, stores it in your account library, and makes it available for any agent in your account.

<Note>
  Cloned voices appear alongside standard voices in the [voice selection](/build/voice-speech/select-voice) panel and can be selected from the **My Cloned Voices** sidebar.
</Note>

<Warning>
  Only clone voices you have explicit written consent to use. You are responsible for ensuring you have the legal right to clone and deploy any voice.
</Warning>

***

## Creating a Cloned Voice

<Screenshot lightSrc="/images/voice-speech__voice-cloning-upload-light.png" darkSrc="/images/voice-speech__voice-cloning-upload-dark.png" alt="Voice cloning form showing provider selection, upload and record tabs, voice name, language, and description fields" />

<Steps>
  <Step title="Open the clone form">
    In your agent editor, go to **General** → **Speaking** and click **Clone Voice**. You can also open it from the **My Cloned Voices** sidebar in the voice selection panel.
  </Step>

  <Step title="Choose a provider">
    Select a cloning provider based on your audio sample:

    | Provider       | Best for                               | Duration                                          | File size       |
    | -------------- | -------------------------------------- | ------------------------------------------------- | --------------- |
    | **ElevenLabs** | Longer samples, more natural variation | Min **5s**, recommended **1-2 min**, max **180s** | Up to **10 MB** |
    | **Cartesia**   | Short clean clips, fast cloning        | Min **3s**, recommended **5-10s**, max **10s**    | Up to **5 MB**  |

    Choose ElevenLabs when you have a longer clean recording and want the model to learn more tone variation. Choose Cartesia when you want a fast clone from a short clean clip. If unsure, create both and compare them.

    **Supported formats:** MP3, WAV, OGG, WebM — ElevenLabs also accepts M4A.
  </Step>

  <Step title="Provide audio">
    Switch between two input modes using the **Upload** and **Record** tabs:

    **Upload** — Drag and drop an audio file or click to browse. The accepted formats and size limit are shown based on the selected provider.

    **Record** — Click the record button to capture audio directly from your microphone. A live waveform visualization and duration counter are displayed while recording. Recording automatically stops when the provider's maximum duration is reached.

    The form tracks the total duration of your audio and shows whether it meets the provider's requirements.
  </Step>

  <Step title="Fill in voice details">
    * **Voice name** (required) — A descriptive name shown in the voice selection panel
    * **Language** (required) — The language of the audio sample (English, Spanish, French, German, Italian, Portuguese, Dutch, Japanese, Korean, or Chinese)
    * **Description** (optional) — Internal notes about this voice
    * **Remove background noise** (optional) — Available for ElevenLabs only — cleans up noise in the sample before cloning
  </Step>

  <Step title="Create the clone">
    Click **Clone Voice** to start processing. The voice status progresses through these states:

    * **Processing** — Provider is generating the voice model
    * **Ready** — Voice is available to assign to agents
    * **Failed** — Something went wrong — check the error message and try again
  </Step>
</Steps>

***

## Managing Cloned Voices

Cloned voices are shared across your account. All team members with appropriate permissions can view and assign cloned voices to their agents.

### Viewing Cloned Voices

Your cloned voices appear in the **My Cloned Voices** sidebar within the voice selection panel. Each voice shows its name, provider, language, and current status badge.

You can also view all cloned voices in a table that shows:

* **Name** and **Provider**
* **Language**
* **Status** (with color-coded badges)
* **Actions** such as selecting or deleting the voice

### Deleting a Cloned Voice

1. Find the cloned voice in the sidebar or table
2. Click the delete button or select **Delete** from the actions menu
3. Confirm the deletion

<Warning>
  Deleting a cloned voice is permanent and cannot be undone. Any agents currently using that voice will need a new voice assigned.
</Warning>

***

## Using Cloned Voices with Agents

Once a cloned voice shows a **Ready** status, assign it to any agent:

<Steps>
  <Step title="Open voice selection">
    In your agent editor, go to **General** → **Speaking**.
  </Step>

  <Step title="Find your cloned voice">
    Look in the **My Cloned Voices** sidebar on the right side of the voice selection panel. Only voices with **Ready** status can be selected.
  </Step>

  <Step title="Select the voice">
    Click the cloned voice to assign it. The agent uses this voice for all subsequent conversations. You can further customize the output with [voice settings](/build/voice-speech/voice-settings).
  </Step>
</Steps>

***

## Audio Sample Best Practices

<AccordionGroup>
  <Accordion title="Recording quality" icon="microphone">
    * Use a good quality microphone (USB condenser or better)
    * Record in a quiet environment with sound dampening
    * Maintain a consistent distance from the microphone (6-12 inches)
    * Avoid rooms with echo or reverb
    * Use 44.1 kHz sample rate or higher
  </Accordion>

  <Accordion title="Content variety" icon="list">
    * Include both short and long sentences
    * Cover different tones: questions, statements, explanations
    * Read naturally at a conversational pace
    * Avoid reading too fast or too slow
    * Include pauses between sentences
  </Accordion>

  <Accordion title="What to avoid" icon="xmark">
    * Background music or ambient noise
    * Multiple speakers in one sample
    * Heavy audio processing, compression, or filters
    * Whispering or shouting
    * Samples shorter than the provider's minimum duration
    * Low-quality phone recordings
  </Accordion>
</AccordionGroup>

***

## Sample Strategy

Use the provider you selected to decide how much audio to collect.

### For ElevenLabs

* Aim for **1-2 minutes** when possible
* Include varied phrasing, not one repeated sentence
* Use the **Remove background noise** option if the recording is otherwise good
* Prefer one speaker, one microphone, one room

### For Cartesia

* Aim for a **short, clean 5-10 second clip**
* Do not over-record just to add more material
* Remove room noise before recording because the clone will reflect it closely
* Choose a clip with stable volume and no interruptions

### Good sample script

Read 4-6 natural sentences in the way you want the agent to sound:

* a greeting
* one short factual sentence
* one question
* one longer explanatory sentence
* one closing sentence

This gives the model enough shape to learn pacing and tone without sounding scripted.

***

## Legal Considerations

<Warning>
  You must have explicit written consent from any person whose voice you clone. Unauthorized voice cloning may violate privacy laws and intellectual property rights.
</Warning>

**Before cloning, ensure you have:**

* Written consent from the voice owner
* Rights to use the voice commercially
* Clear agreement on how the voice will be used
* Documentation of the consent for your records

**Never clone:**

* Voices without consent
* Voices of public figures without licensing
* Voices for deceptive or impersonation purposes

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="My sample is rejected" icon="triangle-exclamation">
    Check the selected provider's duration, file size, and file format limits. Most failed uploads are caused by clips that are too short, too long, or too noisy.
  </Accordion>

  <Accordion title="The cloned voice sounds noisy" icon="wave-square">
    Re-record with less background noise. For ElevenLabs, try enabling **Remove background noise**. For Cartesia, start with a cleaner clip rather than a longer one.
  </Accordion>

  <Accordion title="The voice does not sound expressive enough" icon="sliders">
    Use a better source sample, not just a different speed or pitch setting. Add clearer phrasing variety and natural intonation, then create a new clone.
  </Accordion>

  <Accordion title="I am not sure which provider to use" icon="circle-question">
    Start by cloning the same sample with both providers. Then compare them in the voice picker using the same test script.
  </Accordion>
</AccordionGroup>

***

## Next Steps

<CardGroup cols={2}>
  <Card title="Select Voice" icon="microphone" href="/build/voice-speech/select-voice">
    Browse and compare all available voices
  </Card>

  <Card title="Voice Settings" icon="sliders" href="/build/voice-speech/voice-settings">
    Fine-tune speed, pitch, and stability
  </Card>

  <Card title="Custom Pronunciations" icon="spell-check" href="/build/voice-speech/custom-pronunciations">
    Correct pronunciation for your cloned voice
  </Card>

  <Card title="Test Your Agent" icon="vial" href="/test/web-simulator">
    Test your cloned voice in conversations
  </Card>
</CardGroup>