How a voice clone is made
A voice clone is a speech model conditioned on a recording of one person. Descript's terms define an AI Voice as an AI model trained on voice recordings of a Consenting Speaker, and an AI Speaker as the feature that uses that voice to synthesize new recordings that sound like that person. [2] Once the model exists, you type text and the editor generates audio in that voice, which is why the recording step is also the point where vendors can check consent.
In Descript, you create a speaker label, choose Enable speech generation (or Clone voice to enable AI speech), pick a microphone and record yourself reading the consent statement, then submit. Descript says the speaker is typically ready within minutes. You cannot add more samples to an existing clone: to improve one, you create a new clone with a fresh consent recording. Clones are based on a US English pronunciation model, so accents may not always carry over, and you can switch between ElevenLabs v2, v3 and v4 generation models in App Settings. [1]
In Filmora, the clone is built from the Text to Speech panel: open Modeling Voice, click the plus icon, name the voice, record yourself reading the on-screen text, then click Generate. You can re-record if the result is poor. [5] Neither vendor's guide states a minimum sample length, so the quality advice that is documented is Descript's: speak naturally, with varied tone and expression. [1]
Consent and verification, vendor by vendor
This table compares what each vendor documented on 30 October 2026. Filmora might have checks that its public guide does not describe. We report only what the pages say.
| Question | Descript | Filmora |
|---|---|---|
| How consent is captured | The person records a consent statement during setup, read in English even for non-English output [1] | Guide shows reading on-screen text; no consent statement or verification step is described [5] |
| Who may create the clone | The speaker, or someone acting for them with their recorded authorization; a collaborator can send a recording of the statement [1] | Not stated in the guide [5] |
| Explicitly not allowed | Deceased people, non-consenting people, anyone unable to record the statement, AI or artificial source audio [1]; deceptive or unauthorized consent statements [2] | Wondershare's general terms ban fraudulent or deceptive activity and representing AI output as human-generated [7] |
| Who can see your training audio | Descript staff and contractors may listen to training and synthesized audio to monitor misuse [2] | Not stated in the pages we read |
| Cost | Voice cloning is listed as included on Hobbyist and above; Free is listed as Limited [4] | 100 AI credits per use [6] |
What the terms make you responsible for
Descript's terms (section 9.3) say you will only submit audio of Consenting Speakers who have authorized you to submit it, that you will not read a consent statement on behalf of or imitating another person, and that only you and people you explicitly share access with can generate audio from the voice. Section 8.1 separately prohibits content that impersonates a person and any attempt to clone or imitate the voice of a non-consenting speaker, and section 13 makes you indemnify Descript for claims arising from training audio you submit. [2]
Wondershare's terms of use, which cover its generative AI features, say you are solely responsible for your input, that input containing personal information requires compliance with data protection and privacy laws, and that you must not mislead others about the origin of AI output or remove the watermark that marks it as AI-generated. [7] A voice recording is personal information about the speaker, so the responsibility for consent sits with you even where the app does not ask.
None of this is legal advice. Laws on voice and likeness differ by country and US state, and a vendor policy is a floor, not a full answer.
Sharing, revoking and deleting a cloned voice
In Descript, a speaker can be shared with other members of your Drive through Share voice, and once trained it can be used on any Drive you have access to. The help page does not describe a way to remove a teammate's access short of deleting the voice, and deletion is permanent: deleted voices cannot be recovered. [3] Filmora's cloning guide does not describe sharing or deleting a modeled voice. [5]
That gap matters for teams. Decide before you clone who owns the voice, who may generate with it, and what triggers deletion, such as the person leaving the company or withdrawing consent. Write that down in the consent record rather than relying on app permissions.
Why vendors are strict about this
The US Federal Trade Commission ran a public Voice Cloning Challenge because of the harms it saw. It says families and small businesses can be targeted with fraudulent extortion scams, that creative professionals can have their voices appropriated in ways that threaten their livelihoods, and that policymakers cannot count on self-regulation alone. [8] A cloned voice that can say anything is a liability if it leaks or outlives the consent behind it.
Common situations and what to do
| Situation | Reasonable? | What to do |
|---|---|---|
| Your own voice, to patch misread narration | Yes | Descript: record the consent statement yourself [1]. Filmora: budget 100 credits per clone [6] |
| A teammate who narrates product videos | Yes, with written consent | They record the consent statement themselves; share the voice only with named Drive members [1][3] |
| A podcast guest, to fix one misspoken word | Only if they agree to it explicitly | Ask first, send them the edited line, and consider simply re-recording the line instead |
| A former employee, from old recordings | No, unless they consent again | A clone from archive audio without their participation fails Descript rules [1][2] |
| A celebrity soundalike or a deceased person | No | Descript bans both impersonation and deceased voices [1][2] |
Checklist: cloning a team member's or guest's voice responsibly
- Get written consent that names the tool, the purpose (for example, correcting lines in a named video series), how long the voice will be kept, and who may generate speech with it.
- Have the speaker record the vendor consent statement in person. Never read it for them; Descript prohibits that. [2]
- Tell the speaker that vendor staff may review training audio, where the vendor says so. [2]
- Keep the clone in one workspace and share it only with named people. Record who has access. [3]
- Send every generated line to the speaker for approval before publishing.
- Decide whether to tell viewers that some speech is synthesized. Wondershare's terms already bar presenting AI output as human-generated. [7]
- Agree a deletion trigger and act on it; Descript deletion is permanent, which is what you want when consent ends. [3]
- Store the consent record with the project files so a future editor can see it.
Questions before you decide
Does Descript let me clone someone else's voice?
Only with that person's recorded authorization. The speaker records a consent statement in English, or sends you a recording of it, and Descript will not create clones of deceased people, non-consenting people, or audio from an AI source.
Does Filmora require a consent statement for voice cloning?
Its AI Voice Cloning guide describes recording on-screen text and clicking Generate, with no consent statement or verification step described. Wondershare's terms still make you solely responsible for what you submit.
How much does voice cloning cost in Filmora?
Filmora's AI credits rules list AI voice cloning at 100 credits per use. Generating speech with text to speech then costs 1 credit per 10 characters.
Can I improve a Descript voice clone by adding more audio?
No. Descript says you cannot add samples to an existing clone. To improve it, create a new clone with a fresh consent recording.
Can Descript staff hear my voice recordings?
Descript's terms say staff and contractors may listen to training audio and synthesized audio to monitor for misuse of its voice cloning technology.
Sources & verification
Product details and prices can change. Check the linked provider before buying.
- Descript Help: Create an AI Speaker (voice clone) Accessed 30 Oct 2026
- Descript Terms of Service Accessed 30 Oct 2026
- Descript Help: Manage and share AI Speakers Accessed 30 Oct 2026
- Descript pricing Accessed 30 Oct 2026
- Filmora guide: AI Voice Cloning Accessed 30 Oct 2026
- Filmora AI credits rules Accessed 30 Oct 2026
- Wondershare Terms of Use Accessed 30 Oct 2026
- FTC: Voice Cloning Challenge Accessed 30 Oct 2026
Sources link directly to providers. Product buttons may use separately labeled affiliate links. Read our disclosure.