Generally available, with a gate in front of it

The launch wording deserves a slow read. "Generally available" usually means any customer can open the console and use the feature. Here, access is already enabled for select Amazon Polly customers, and others can reach out to their AWS account team or apply for access. That is a defensible way to ship voice cloning — the misuse case is obvious — but it changes what the announcement is. Teams planning a product around a cloned brand voice cannot schedule around this launch until their account is switched on, and AWS gives no timeline for wider enablement.

What the terms actually give you

Building the voice costs nothing, and the input ask is small: minutes, not hours, of one person's speech. Using the clone is billed pay-as-you-go; the launch post points to the Polly pricing page and gives no rate of its own. The constraints sit elsewhere. Ten voices per account per Region is the initial quota; increases go through the account team or AWS Support. A voice exists only in the account and Region where it was created; to serve another Region or account, you create the voice again there, from the same recordings, and the speaker's consent must cover every copy. And the language binding cuts against the obvious localization pitch: cloning a spokesperson in English does not produce their Spanish voice. Seven languages are supported, but each one is a separate clone, recorded and reviewed in that language.

Read the safeguards as a list: a consent confirmation in the API call, a watermark on generated speech, source recordings not retained after creation, no training on customer recordings, IAM access control, and optional encryption with the customer's own KMS key. During creation, Polly also checks the new voice for speaker similarity. What the setup cannot do is verify the consent itself. TermsAccepted records the customer's own confirmation, given for each voice created. Whether the person in the recordings agreed is still the customer's word, and AWS's own guidance tells customers to keep a record of each speaker's consent and to delete the voice if consent is withdrawn.

Who this is really for

The fit is narrow and clear: organizations that already run on AWS, already use Polly's generative voices, and need one consistent voice they control — service prompts that change weekly, narration that gets revised after recording, accessibility reading in a familiar voice. For them, a free creation step and pay-as-you-go usage inside an existing bill is a straightforward trial, once the gate opens. For a team choosing a vendor this week, the deciding facts are the gate, the one-language-per-clone limit, and the account-and-Region lock — all stated limits in the launch post, not roadmap promises.

Sources

  1. [1] AWS Machine Learning Blog — "Amazon Polly launches voice cloning: A unique voice for your brand," November 19, 2026 (primary; launch terms, languages, input limits, consent, watermark, retention and scoping)Read source
  2. [2] Amazon Polly API documentation — SynthesizeSpeech / CreateVoice custom-voice behavior, including vc- voice identifiers and active-status requirementRead source
  3. [3] Unite.AI — "Amazon Polly Adds Gated Voice Cloning for Custom Brand Voices," November 19, 2026 (secondary cross-check of gated access, Regions, quota and recording guidance)Read source