Oossa

AWS launches gated voice cloning in Amazon Polly

Amazon Polly can create a synthetic voice from a few minutes of a consenting speaker’s recordings. Access is limited to enabled AWS accounts, and usage is paid.

By Published by Oossa: 2 min read

Amin Asbaghipour · Unsplash

AWS has made voice cloning generally available in Amazon Polly, but access is gated by account. The feature creates a synthetic voice from recordings of one speaker who has agreed to be cloned. Businesses can then use that voice to generate speech for phone systems, apps, games, lessons and other content, without recording every new line in a studio.

AWS says voice creation is free, while generating speech with a cloned voice costs money on a pay-as-you-go basis. The feature launches in English, Spanish, German, French, Italian, Portuguese and Hindi. A cloned voice speaks only the language used in its recordings; AWS says accents beyond its tested recording locales are less tested.

How the cloning process works

Customers submit one to 10 recordings, with a total limit of 15 minutes. AWS recommends two to 10 minutes of natural speech, recorded by one speaker in a quiet room. Creating a voice usually takes five to 10 minutes and happens in the background. Once it is ready, developers can use it through Polly’s speech-generation tools, including for longer audio saved to Amazon S3.

The feature is not automatically enabled for every customer. AWS says select Polly customers already have access; others can ask their AWS account team or apply. It is offered in the AWS Regions that support Polly’s generative engine, including the US GovCloud region. The initial limit is 10 cloned voices per account in each Region.

Consent and control are part of the setup

When creating a voice, the customer must confirm they have the speaker’s consent. AWS says it adds a watermark to speech generated with cloned voices, deletes its copy of the source recordings after voice creation, and does not use those recordings to train Polly models. The resulting voice data is encrypted and remains until the customer deletes it.

A cloned voice is available only to the AWS account and Region where it was created. Account owners can control which users and services can use it, and can delete it. To make a voice available in another Region or account, customers must create it again from the recordings and ensure the speaker’s consent covers each copy.

AWS positions the feature for uses such as consistent customer-service prompts and narration that can be revised without booking another recording session. But it advises customers to listen to the voice with their own scripts before using it: names and numbers may need testing, and the same text can sound slightly different across requests.

Why it matters

For companies using Polly in customer-service lines or apps, the feature could let them update spoken prompts while keeping a consistent voice. Access is gated, and businesses still need the speaker’s consent and must test generated speech with their actual scripts before using it.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.