> ## Documentation Index > Fetch the complete documentation index at: https://cntillion.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Voices > Find, preview and create voices for every agent in your workspace. Open **Voices** in the workspace sidebar. **Explore** contains the voice catalog; **My voices** contains the voices your team has saved or created. The same library is available to every agent in the workspace. ![Workspace voice library with search, filters and voice previews](https://mintcdn.com/centillion/Aanjhz7JPJP5Ijzk/images/workspaces/voices-01-library.webp?fit=max\&auto=format\&n=Aanjhz7JPJP5Ijzk\&q=85\&s=47a7cf775797b2c2bb96ada184b72ce8) Explore voices by language, accent and speaking style, then listen before choosing. ## Find a voice Search by name or description, choose a language, or pick a use case such as Conversational or Narration. **Filters** includes quality, gender, age, accent, speaking style, pitch and notice period. You can also include voices with live moderation. Available details vary by voice. Voices with custom generation rates are excluded from the catalog. Sort by Trending, Latest, or Name A to Z. **Load more** continues the results. **Clear all filters** resets your choices. The microphone button beside search opens **Find similar voices**. Upload or record a sample to find public voices with a similar sound. This does not clone the speaker. Press a voice's play button to listen. The player supports pause, replay and seeking. Choose **Edit sample** in the player to change the sample text, then press play. Previews are available on every plan; saving, creating and assigning voices requires voice access. ## Save and assign Save a voice to add it to **My voices**. Choose **Use voice**, select an agent, and save the assignment. You can also select the current voice under **Agent → Configuration → Voice**. Publish the agent's branch before expecting changed speech settings in new calls. A voice assigned to an agent cannot be removed until that agent uses another voice. Removing a voice removes its saved library entry. It does not delete the original recording or trained voice account asset. ## Create a voice Choose **Create voice**, then pick how to create it. ![Create voice dialog with instant clone, professional clone, voice design and voice remixing](https://mintcdn.com/centillion/xsam3k7c8Fik7_oP/images/workspaces/voices-02-create.webp?fit=max\&auto=format\&n=xsam3k7c8Fik7_oP\&q=85\&s=664c2c289d98e7f945162e1fd1bf0b24) Create a clone, design a new voice, or remix one you have saved. ### Instant voice clone Upload or record at least 10 seconds of clear audio. Use a quiet room, a good microphone and consistent speaking volume. Each file can be up to 10 MB. Depending on the available cloning option, you may be asked for a single sample between 10 and 60 seconds. ![Instant clone upload step with recording guidance and upload or record controls](https://mintcdn.com/centillion/xsam3k7c8Fik7_oP/images/workspaces/voices-03-clone.webp?fit=max\&auto=format\&n=xsam3k7c8Fik7_oP\&q=85\&s=77ff1b19d9d9da8e5deb10cd848f25d2) Upload a sample or record one, then enter the voice details and confirm permission. Choose **Next**, enter a voice name, language and optional description and labels. Confirm that you have permission to use the speaker's voice. Review the details and create the voice. A voice requiring additional verification remains disabled in **My voices** until verification is complete. ### Professional voice clone Use this for a detailed replica of your own voice. Supply at least 30 minutes of usable audio; two to three hours is recommended. Each sample can be up to 100 MB. Keep the microphone, recording environment and speaking style consistent. After uploading, return to the voice in **My voices** to finish setup. Separate speakers when a sample contains more than one voice, listen to them, and select your speaker. Only selected audio counts toward the training minimum. Complete the spoken verification challenge when requested, then start training. If spoken verification fails, **Manual verification** lets you submit the ownership documents requested by support. You can attach up to five PDF, PNG or JPEG files, up to 10 MB each, and add a note. Training and review take time; refresh the saved voice to check progress. The voice becomes usable when training is complete. ### Voice design Describe a new voice's tone, accent and delivery. Optionally enter preview text, or let Centillion Labs generate a script. **Advanced settings** contains voice style, prompt guidance, loudness, a seed and description expansion. Generate previews, listen to the alternatives, enter a name and choose **Use this voice** to save the one you prefer. ### Voice remixing Choose an eligible saved voice and describe how it should change. Generate previews and save a variation with a new name. Remixing creates a separate voice; your original voice remains available. Only voices that support remixing appear in the selector. ## Uploads and access Samples and ownership documents use private temporary storage. Files are deleted after transfer for processing; abandoned uploads are removed after 24 hours. Only authorized workspace members can create, save, remove or assign voices. ## Voices and call mode An agent's **Voice** configuration selects Realtime or Custom. The voice picker lists every voice. Those the current mode can't use read **Use in Custom** or **Use in Realtime**, and choosing one switches the mode. Realtime uses the provider's own voice controls; Custom exposes supported speech tuning. Changing a voice does not replace the selected Custom language model. See [Voice mode](https://cntillion.com/docs/agents/create-an-agent.md#voice-mode) before changing an agent that transfers calls or uses structured procedures. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.