Responsible use
loudkit clones a voice from a few seconds of audio and reads arbitrary text in it, locally, with no account and no network. loudkit cannot verify who owns a voice sample. That is your responsibility, and it does not transfer to us.
Allowed
Section titled “Allowed”- Your own voice, or a voice you have written permission to use.
- Public-domain and openly licensed recordings, within their terms.
- Accessibility: screen reading, assistive speech, restoring a voice someone has lost.
- Narration, localisation, dubbing, games, and research, with the consent of whoever the voice belongs to.
- The shipped voices are enrollments of recordings made or released for speech-technology use: personal donations recorded expressly for TTS (CC0), and CC0 / CC-BY corpora whose terms allow synthesis. docs/voices/roster/provenance.json names the donor or source, licence and consent basis per voice. No anonymous scraped audio ships here, and no voice this project cloned from a private individual ships either.
Prohibited
Section titled “Prohibited”- Impersonating a real person to deceive, defraud, or harass.
- Putting words in a public figure’s mouth, including satire that is not clearly labelled as synthetic.
- Defeating voice authentication, or helping anyone do so.
- Cloning a voice from a recording published for another purpose (a podcast, a lecture, a video) without the speaker’s consent. Public is not consenting.
- Distributing a voice pack of an identifiable person who has not consented. Publishing one clip is not the same act as publishing a file that reproduces someone’s voice on demand, for everyone who downloads it.
Disclosure
Section titled “Disclosure”Label synthetic audio as synthetic, and cite the source recording when you publish a sample.
On the law. The EU AI Act’s transparency obligations for synthetic content sit in Article 50. It requires providers of systems generating synthetic audio to mark the output in a machine-readable way. It does not name a format, and it does not name C2PA. The Commission’s transparency guidelines cover which obligations fall on whom, plus the exemptions and transitional arrangements. C2PA Content Credentials are a widely used way to satisfy a machine-readable-marking requirement. They are a choice this project made, not a compliance verdict it can hand you.
None of this is legal advice, and shipping loudkit’s manifest is not a finding that you comply. Whether you are a provider or a deployer, and what that obliges, depends on facts about you. Ask someone qualified.
loudkit writes that marking by default. Every saved WAV (Result.save) and
every server response carries a claim-only C2PA manifest (a JUMBF c2pa
box, plus the X-Loudkit-Provenance header over HTTP) with the algorithm
fingerprint, the seed, and the SHA-256 binding it to the audio bytes. See
docs/reference/provenance.md. It is unsigned,
because signing needs a certificate and that is the deployer’s choice. It says
what made the file, not who vouches for it.
This is the machine-readable marking loudkit provides by default. Deployers can add their own signing policy and disclosure around it.
What we ship, and what we will not
Section titled “What we ship, and what we will not”We ship twenty voice profiles enrolled from recordings donated for speech technology or released under terms that permit this use. Every profile has a named source, licence and consent basis in the public roster. We do not ship profiles made from private recordings or recordings published for an unrelated purpose without the speaker’s permission.
The enrollment code is included so you can make your own profile from a voice you have the right to use.
Issues and pull requests asking for help with undisclosed impersonation, voice authentication bypass, or stripping provenance from generated audio will be closed.