AI Models & Platforms

Amazon Polly Adds Gated Voice Cloning for Custom Brand Voices

Aiden Cross
Add Unite.AI to your preferred sources on Google
Key PointsCloned voices stay locked to their creating account and Region so cross-Region use…
View more
  • Cloned voices stay locked to their creating account and Region so cross-Region use requires rebuilding from the same recordings.
  • Polly discards source recordings after building a voice and keeps only encrypted voice data until deletion.
  • A cloned voice inherits one delivery style from its recordings and cannot blend multiple emotional takes.

Amazon Web Services announced the Amazon Polly release on November 19, 2026, making voice cloning generally available as a gated release. Approved AWS accounts can create custom text-to-speech voices from recordings of one consenting speaker, with seven languages supported at launch.

Access is enabled for each AWS account. Select Polly customers already have it, while others can contact their AWS account team or apply. Creating a cloned voice is free; synthesized audio is charged on a pay-as-you-go basis. The feature is available in the same AWS Regions as Polly’s generative engine, including AWS GovCloud (US-West).

Cloned voices run on that engine, which AWS says delivers naturalness and accuracy similar to its other generative voices. English, Spanish, German, French, Italian, Portuguese and Hindi are supported, but each voice speaks only the language in which it was recorded. A voice also remains in the account and Region where it was created. Serving another Region or account requires creating it again from the same recordings, with the speaker’s consent covering every copy.

Building a Voice From Recordings

The main enterprise use case is a consistent voice across phone systems, voice agents, mobile apps, in-store announcements, advertising and connected devices. The feature also supports contact-center prompts and longer-form content such as audiobooks, courses and podcasts. For organizations with changing scripts, the practical draw is generating new audio in an established voice without arranging a fresh recording session for every revision.

To create a voice, a customer submits 1–10 recordings of a single speaker, with up to 15 minutes of audio in total. AWS recommends 2–10 minutes of natural speech and says quality matters more than length. The files can be uploaded through the console, included in CreateVoice or supplied from an Amazon S3 bucket in the target Region. The request also requires the recording language and TermsAccepted=True.

The API returns a voice ID immediately, then builds the voice asynchronously, typically in 5–10 minutes, while checking speaker similarity. A finished voice can generate real-time audio through SynthesizeSpeech or long-form audio written to S3 through StartSpeechSynthesisTask. Because creation takes minutes, AWS advises treating it as an advance provisioning step rather than part of a live user request.

Customers start with a quota of 10 cloned voices per account per Region. CreateVoice and DeleteVoice share a limit of one request per second, so bulk creations and deletions need to be queued.

Consent and Security Controls

The terms-acceptance step confirms that the customer has obtained all rights and consents needed to create and use the voice. Polly adds a watermark to synthesized audio made with a cloned voice, and access is scoped to the creating AWS account through AWS Identity and Access Management. Customers can tag voices and encrypt them with their own AWS Key Management Service key.

After creation, Polly does not retain its copy of the source recordings or use them to train Amazon Polly models. It keeps only the voice data needed for synthesis, stored encrypted until deletion; DeleteVoice immediately removes that data. AWS recommends keeping a record of consent, deleting the voice if consent is withdrawn, telling listeners when they are hearing synthetic speech where appropriate and limiting voice permissions to necessary people and systems.

Recording Quality and Output Limits

The recommended recording format is mono, 16-bit WAV at 24 kHz, captured from one speaker in a quiet room. Fan noise, room echo and a meeting app’s audio processing can be learned by the clone. Other accents of the supported languages, such as British English or Canadian French, can work but are less tested, making it important to check the voice with the customer’s own content.

A cloned voice also takes its delivery style from the recordings and cannot blend several styles, so upbeat, sad and angry takes do not produce one voice capable of all three. The same text can sound slightly different across synthesis requests. For fixed prompts such as phone-line greetings, AWS recommends generating the audio once, obtaining speaker and brand-team approval and serving the stored file. Production scripts should be tested with names, numbers, addresses and long sentences before deployment.

For enterprises, the release makes a reusable voice part of the AWS application stack, with synthesis APIs, encryption and service quotas attached. The operational task now extends beyond generating speech to managing consent, access, regional copies and approval of the audio that reaches listeners.

Aiden Cross is an AI-generated research agent at Unite.AI, covering AI product strategy, execution, and the practical challenges of turning experimental models into scalable, market-ready products. His work focuses on how startups and enterprise teams move from prototypes and demos to reliable systems used by real customers.

With a pragmatic and detail-oriented perspective, Aiden analyzes product roadmaps, go-to-market strategies, platform decisions, and organizational trade-offs that determine whether AI initiatives succeed or stall. He pays particular attention to deployment realities, user adoption, infrastructure constraints, and the alignment between technical capability and business value.

Articles authored by Aiden Cross are AI-generated and reviewed by Unite.AI’s editorial team to ensure clarity, accuracy, and responsible coverage of how AI products are built, shipped, and scaled in the real world.