From the source
Amazon Web Services made Amazon Polly voice cloning generally available as a gated release, letting approved AWS accounts build custom text-to-speech voices from recordings of one consenting speaker.
- Access is enabled per AWS account, and select Polly customers already have it while others can contact their AWS account team or apply.
- Customers submit 1–10 recordings of a single speaker, up to 15 minutes of audio in total, and AWS recommends 2–10 minutes of natural speech, saying quality matters more than length.
- The API returns a voice ID immediately and builds the voice asynchronously, typically in 5–10 minutes, so AWS advises treating creation as an advance provisioning step rather than part of a live user request.
- Seven languages are supported at launch, but each voice speaks only the language in which it was recorded, and a voice stays in the account and Region where it was created, so serving another Region or account requires rebuilding it from the same recordings.
- Polly adds a watermark to synthesized audio from a cloned voice, discards its copy of the source recordings after building the voice, and keeps only encrypted voice data until deletion.