Better voice cloning raises the question of whose voice it is
Improved voice cloning makes it easier to produce narration in many languages, and it makes the question of whose voice you are using more important. ElevenLabs’ September 28, 2026 changelog describes Eleven v4 as offering higher-quality, more expressive speech and improved voice cloning in more than 90 languages, aimed at content creation and long-form audio.
A creator can now imagine a single narrator voice carrying a tutorial series into several languages. That is useful, and it depends on a voice that someone owns and has agreed to share. A short record kept with each cloned voice makes that agreement visible to you, your collaborators, and later to anyone who asks.
The record in this article has four lines: the voice, the consent, the source audio, and the use. It is editorial synthesis, not an ElevenLabs feature or a consent form. It differs from the voice-cloning aspect of AI dubbing release checks because it starts earlier, with the decision to create the voice.
Know what the sources say and what they leave open
The changelog also describes Eleven v4 Turbo, tuned for real-time use, and notes that Text to Dialogue requests can use the Instant Voice Clone version of a Professional Voice Clone. It does not give pricing for Eleven v4, plan or regional availability, or what happens to older models.
ElevenLabs’ voice cloning documentation describes two kinds of clone. Instant Voice Cloning approximates a voice from a short sample without training a custom model, and Professional Voice Cloning trains a dedicated model on a larger dataset and requires a verification process to confirm the voice belongs to you. Professional clones need a Creator plan or higher.
The documentation does not describe commercial use rights or licensing for generated audio, any requirement to disclose that audio is AI-generated, or the exact wording of a consent confirmation. Those are questions for ElevenLabs’ terms and for you, so do not assume the documentation answers them.
Start with whose voice it is
The documentation says you can only create a Professional Voice Clone of your own voice, even with someone else’s consent. A person who wants to share their voice can create and verify a Professional Voice Clone on their own account and then share it with you through a sharing link.
That route matters for creators who work with guests or contractors. Rather than collecting recordings of another person and cloning them yourself, the person who owns the voice makes the clone, and you receive access to it. The record should say which route you used.
For a hypothetical example, imagine an independent educator who wants one consistent narrator for a tutorial series in several languages. If the narrator is the educator, they can verify and use their own clone. If it is a colleague, the colleague creates and shares it. The documentation states its rule for Professional clones, so check what applies to any other kind before you consider cloning a colleague’s voice from old recordings, and get their agreement first either way.
Write down what the person agreed to
Consent for a cloned voice is more than a yes to being recorded. Write down the projects the person approved, the languages, the channels where the narration will appear, how long the permission lasts, and how they can withdraw it. Keep it in writing and attach it to the record.
The documentation does not describe the form of a consent confirmation, so create your own. A short message that states what you plan to do and a reply that agrees is better than memory. If the voice belongs to someone else, ask them to review the list of languages, since a clone may be used in languages the person does not speak.
If the plan changes, such as a new channel or a sponsored use, ask again. A permission given for tutorials does not automatically cover advertising.
Keep a note on the source audio
The documentation’s principle is that consistent, good input leads to consistent, good output. It describes using one speaker with no background noise, music, or echo, keeping volume, tone, and delivery consistent, and choosing clean clips over more length. It gives different amounts of audio for the two clone types.
Keep a short note of which recordings you used, who made them, and where they came from. If you use audio that belongs to someone else, such as a conference recording, confirm that you have the right to use it for this purpose.
This note also helps you troubleshoot. If a narration sounds inconsistent in a particular language, you can look at what the clone was built from before you blame the model.
Decide where it appears and how you will say it is synthetic
List the channels and languages where the cloned voice will be used. A voice approved for a tutorial video should not quietly appear in an ad, a podcast, or a real-time agent. Ask the voice owner again for anything outside the original list.
The documentation does not say whether you must disclose that audio is AI-generated, so decide your own approach and check the platform rules where you publish. The AI disclosure workflow and the disclosure carryover matrix cover how to classify and carry a disclosure across formats.
Before you publish, listen to the result in each language, ideally with someone who speaks it. A cloned voice may sound like the owner but mispronounce a term or stress a word wrongly, and the owner’s name is on that sound.