Skip to content

Responsible use

Voice cloning can impersonate people. The pipeline treats that risk as a design requirement.

Consent is required

Custom-voice requests must include an explicit consent flag, or they are rejected. Catalogue voices record the speaker’s permission.

Inaudible watermark

Every generated chunk carries Resemble AI’s Perth watermark, which survives typical editing and identifies the audio as synthetic.

Labelled files

Every WAV and MP3 is tagged “AI-generated synthetic voice” in its metadata, and the studio shows an AI badge on every result.

Privacy

Client IPs are stored only as salted hashes, uploads are validated and size-limited, and the API never exposes file paths.

Why the samples on this site use a built-in voice

Section titled “Why the samples on this site use a built-in voice”

The pipeline was developed and measured on a client-provided reference voice. That recording is authorised for local studio use only, not for public publishing. So:

  • Measurements from the client voice are reported as numbers only.
  • Every audio sample on this site uses the model’s built-in demo voice, which isn’t a real person.