Using Synthetic Voices Without Deceiving Anyone

Why a voice is not like other content

A voice is part of someone's identity. People recognize friends from a single word on the phone, trust a familiar presenter, and some banks have used voice recognition to verify customers. A photo shows what someone looked like at one moment; a voice model can make them say anything, in any language, indefinitely.

That is why the harms of misused voice technology land on a person rather than a file. If a synthetic version of a teacher, a chief executive or a politician says something false, the damage attaches to the real human, and corrections rarely travel as far as the original clip.

Consent that actually means something

Consent is the foundation of ethical voice synthesis, and a click-through checkbox rarely meets the bar. Meaningful consent has several properties:

• Informed: the person understands that synthesis can produce speech they never recorded, in languages they may not speak.
• Specific: it names the uses, such as product videos in Spanish and Japanese for two years, rather than any purpose forever.
• Revocable: there is a clear way to withdraw, and a plan for what happens to existing content when they do.
• Fair: commercial use of a person's voice is usually paid, separately from any other work they did.
• Documented: written down and stored where someone can find it later.

Some situations need extra care. Employees may feel unable to refuse a manager's request, so make refusal genuinely safe. People who have died cannot consent; their families or estates may have views and, in some places, legal rights. Minors need a guardian's involvement. Being famous is not consent: a public figure's voice is not raw material.

Disclosure and labeling

Listeners assume a voice reflects a real person speaking unless told otherwise. Telling them otherwise is cheap and preserves trust. A note in the video description, a line on screen at the start, or a clear name on an alternate audio track, such as "Spanish (AI voice)", all work.

Several major video platforms now ask creators to disclose realistic synthetic or altered content, and some add labels themselves; the details differ and change, so read each platform's current help pages. Technical provenance is developing too: some voice models embed imperceptible watermarks, and content credential standards aim to record how media was made. Neither replaces a plain-language label, because most viewers never inspect metadata.

Impersonation, fraud and deception

The clearest ethical line is impersonation. Synthetic voices have been used in family emergency scams, fake instructions from executives to finance staff, political robocalls and endorsements celebrities never gave. In each case the deception depends on the listener believing a specific real person is speaking.

The rule that follows is simple: never make an identifiable person appear to say something they did not say, unless they have agreed to that exact use. Satire and parody sit in a gray zone; at minimum, make the artificial nature unmistakable, because a clip will be shared without its context.

Dubbing someone else's words

Translation dubbing raises a quieter issue. A synthetic voice reading a translation of a real speaker presents those words as theirs. If the translation is wrong, the speaker has been misquoted in a language they may not understand and cannot correct. That makes accuracy an ethical obligation, not just a quality goal, especially for medical, legal, safety or HR content.

It also matters whose video it is. Dubbing a film, a course or another creator's upload without permission is a rights problem before it is an ethics problem. And using a neutral preset voice rather than a clone of the speaker is a useful signal in itself: it tells viewers they are hearing a translation, not the person.

The effect on voice actors and translators

AI voices are already taking some work that used to go to people, especially routine narration, e-learning and low-budget localization. Voice actors have also raised concerns that recordings were used to train models without clear consent or payment. Performer unions have negotiated consent and compensation terms for digital replicas in some contracts, and many actors now read licensing agreements closely for AI clauses.

There is another side. A lot of video was never going to be dubbed by humans at all, because the budget did not exist; for that content, an AI voice adds access rather than replacing a job. Being honest about which case you are in is part of using the technology responsibly. Hybrid workflows, with AI drafts reviewed by translators or high-value content voiced by actors, keep people in the roles where judgment matters most. AI dubbing vs human dubbing compares the two in practical terms.

Risks and gray areas that remain

• Labels get ignored or stripped when clips are reposted, so disclosure reduces deception but cannot guarantee it.
• Watermarks can be weakened by compression or deliberate editing.
• Users of a voice product usually cannot verify how the underlying model's training data was gathered.
• Accessibility and authenticity can pull in different directions; a translated voice helps viewers who cannot read subtitles, even though it is not the speaker's own.
• Laws on deepfakes, publicity rights and transparency vary by country and are changing. This article is not legal advice.

How mydubly approaches voice ethics

mydubly's AI dubbing offers eight preset voices only. It does not clone voices, and users cannot upload anyone's recording to create a voice, which removes the most direct route to impersonating a specific person. One voice reads the whole video, so a dub presents itself as a translation rather than an imitation of any participant.

Every full translation also returns transcripts in both languages, which lets a fluent reviewer check exactly what the voice says before anything is published. On privacy, the video file stays on your device; only the audio is uploaded, and uploaded audio and results are deleted within 30 minutes of a job finishing.

The acceptable use policy sets the rules: process only content you own or have permission to use, and do not use translated voices or transcripts to impersonate a real person, mislead people about who said what, or create deceptive media such as political disinformation or fraudulent endorsements. It also recommends labeling dubbed content as machine-translated.

A responsible-use checklist

1. Confirm you have the right to translate the video and, where a person is speaking, that they are comfortable with a translated version.
2. Use a preset voice unless the speaker has given written, specific consent to a clone.
3. Have someone fluent in the target language review the translated transcript, prioritizing anything that could cause harm if wrong.
4. Add a plain disclosure in the description, on screen or on the audio track name.
5. Link to the original so viewers can check the source.
6. Follow each platform's current rules on synthetic media.
7. Fix and republish quickly if someone reports a mistranslation.

Where to go next

If your project passes that checklist, produce the dub with mydubly's AI dubbing tool and keep the transcripts with your published files as a record of what was said. For the practical side of choosing between a clone and a preset voice, read voice cloning vs stock voices.