Voice Cloning.
Voice cloning is the reproduction of a specific person's voice with AI. From recordings, in some systems just a few seconds, a model learns timbre and speaking style and can then speak any text in that voice. It is only permissible with the consent of the person whose voice is cloned.
Voice Cloning — Explained in Detail
Voice cloning builds on text-to-speech. Instead of a ready-made standard voice, the system uses a voice learned from recordings of a real person. Broadly, there are two approaches: fast cloning from short recordings produces a recognisable but imperfect copy; more elaborate training with longer, clean recordings yields a voice that also captures stress and idiosyncrasies well. Both are now available as cloud services and as open-source models for local use.
There are several sensible uses: a managing director voices explainer videos in her own voice without going into a studio each time. A company keeps the voice of its phone announcement consistent even when the texts change. People at risk of losing their voice through illness have it preserved beforehand. What all these cases have in common: the person whose voice is cloned knows about it, has consented and can control how it is used.
The downside is obvious. Cloned voices are used for fraud, for example in calls in which a family member or the boss supposedly needs money urgently. Companies should therefore never base payment approvals on a call or voice message alone and should have internal verification rules. Transparency obligations increasingly apply to published content with artificially generated voices, in the EU for example under the EU AI Act for deepfakes. Anyone using AI voices in advertising or videos is best advised to label them openly.
Legally, in Switzerland the voice is part of a person's personality. Article 28 of the Swiss Civil Code (ZGB) protects against unlawful infringement of personality, and reproducing and using a voice without consent can constitute such an infringement (text: fedlex.admin.ch, SR 210). Voice recordings are also personal data under the Data Protection Act (SR 235.1); if they are processed to identify a person uniquely, they count as biometric and therefore sensitive data. In practice this means written consent with a clear purpose, duration and option to withdraw. This overview is not legal advice.
What proper consent covers
Consent for voice cloning should record in writing: whose voice is cloned, what it may be used for (for example only the company's explainer videos, no advertising for third parties), for how long, where the recordings and the voice model are stored, who has access and how the person can withdraw consent. What happens to the voice after someone leaves the company belongs in it too.
Technically, running locally is often the cleaner solution because recordings and the voice model never leave the premises. How we plan AI projects with data protection in mind is described on AI consulting Switzerland; basics on AI in companies are on AI for SMEs.
Example: a driving instructor and his learning videos
A constructed example: a driving instructor in Aargau produces short learning videos for the theory test. He wants all videos to be spoken in his voice but does not have time to re-record every correction. He records half an hour of clean text and has a voice model created locally from it. New videos are now produced from the script.
Because it is his own voice, he needs no one else's consent, but he records in writing for his employee that her voice will not be cloned. Each video's credits state that the voice is AI-generated. The voice model is stored only on his computer so that it cannot be misused for other purposes.
Related Page
Text-to-Speech (TTS)Frequently Asked Questions About Voice Cloning
With the consent of the person concerned, generally yes. Without consent, reproducing and using a voice can infringe personality rights (Art. 28 Civil Code), and voice recordings are subject to the Data Protection Act. If in doubt, and for commercial use, seek legal advice.
That depends on the system. Some produce a recognisable copy from a few seconds; a truly convincing voice requires longer, clean recordings without background noise. The better the source material, the more natural the result.
Never base payments or sensitive approvals on a call or voice message alone. Call-backs to known numbers, a four-eyes principle for transfers and internal code words work well. Staff should know that voices can now be reproduced deceptively well.
More Terms Starting with "V"
Voice Search Optimization
Voice Search Optimization is the adaptation of web content for spoken search queries via voice assis...
Vector Embedding
A vector embedding is the translation of text into a sequence of numbers that represents its meaning...
Vibe Coding
Vibe coding describes a way of working in which a person describes the desired function in ordinary ...
Ready for Your Project?
Apply this knowledge to your website — DLM Digital will help you.