Amazon Web Services has launched voice cloning in Amazon Polly, its text-to-speech service. The feature creates a synthetic voice from recordings of one speaker, then uses that voice to read new text. AWS calls the launch generally available, but access is gated and currently enabled for select Polly customers.
Voice cloning supports English, Spanish, German, French, Italian, Portuguese and Hindi. Each cloned voice speaks only the language used in its recordings. AWS says voice creation typically takes five to ten minutes.
Customers create voices from recorded speech
Customers can submit one to ten recordings through the Polly console or its CreateVoice API. AWS recommends two to ten minutes of speech and allows up to 15 minutes of audio per voice. A customer must confirm that the speaker has consented each time they create a voice.
Once a voice is ready, customers can use it to generate speech in real time or create longer audio files. AWS describes uses including contact center prompts, narration, game characters and assistive reading tools. It says a cloned voice takes its speaking style from the recordings and cannot combine different styles.

AWS limits access to each cloned voice
AWS says each voice is available only to the account that created it, with access controlled through AWS Identity and Access Management. Generated speech carries a watermark. The company says it does not retain the source recordings after creating a voice or use them to train Polly models. It stores the voice data needed to generate speech in encrypted form and says deleting a voice removes that data immediately.
Creating a voice is free for Polly customers, while use is charged on a pay-as-you-go basis. AWS did not give a usage rate in its announcement. Voice cloning is available in the AWS Regions that support Polly’s generative engine, including AWS GovCloud (US). Customers without access can contact their AWS account team or apply for it.



