Skip to content
trendoviz

Amazon Polly adds gated voice cloning

Amazon Polly voice cloning is now generally available as a gated release, turning a consenting speaker’s recordings into a custom text-to-speech voice. AWS says creation typically takes 5–10 minutes, with voice use billed on a pay-as-you-go basis.

51 out of 100Emerging7D: Insufficient history ProvisionalPublished

Overview

Amazon Polly voice cloning is now generally available as a gated release. Access is enabled for select Amazon Polly customers; others must contact their AWS account team or apply for access [1]. The feature is available in all AWS Regions supported by the Amazon Polly generative engine, including AWS GovCloud (US) [1].

The release supports English, Spanish, German, French, Italian, Portuguese, and Hindi. Each cloned voice speaks only the language in which it was recorded, so multilingual services require separately created and approved voices [1]. AWS describes the cloned voices as running on the Polly generative engine, with naturalness and accuracy similar to its other generative voices; that performance comparison is a vendor claim rather than an independently tested result [1].

Why it matters

Voice cloning changes how teams produce scripted audio. AWS presents the feature as a way to generate on-brand audio on demand and update prompts without rebooking a recording session each time a script changes [1].

That workflow introduces operational work beyond the synthesis API. Teams need processes for recording approval, consent confirmation, voice creation, quality review and distribution of generated audio. Because each voice is tied to one recording language, multilingual deployments also require a voice library, review workflow and consent process for each language [1].

The feature is one component of a broader voice-agent stack rather than a standalone text-to-speech product: AWS specifically lists cascaded voice agents among its proposed applications [1].

Opportunities

Developers can use cloned voices in interactive voice-response systems, support and self-service flows, and voice-agent applications [1].

Businesses can maintain one recognizable voice across channels such as mobile apps, in-store announcements, advertising and connected devices, according to AWS’s stated use cases [1].

Creators and studios can produce audiobooks, course narration, podcast segments, game-character dialogue and video narration from approved source recordings. AWS says creators can produce later chapters or updates quickly and maintain a coherent voice identity across a content library [1].

Risks

Consent and rights: The creation flow requires TermsAccepted=True, confirming the speaker’s consent, and AWS says consent must be confirmed each time a voice is created [1]. The evidence does not establish whether a speaker’s permission covers the intended duration, channels, languages, territories or commercial uses.

Language coverage: A cloned voice reads only in the language in which it was recorded. Multilingual services therefore need separate approved voices for each supported language [1].

Quality review: AWS checks the created voice for speaker similarity during the asynchronous build, but the evidence does not say what happens if a voice fails that check [1].

Context

The creation workflow begins with recordings from one consenting speaker. A CreateVoice request includes 1–10 recordings, the recordings’ language, and TermsAccepted=True, confirming that the speaker has consented to voice cloning. Amazon Polly returns a voice ID immediately and builds the voice asynchronously [1].

AWS says creation typically takes 5–10 minutes. During that period, Polly checks the resulting text-to-speech voice for speaker similarity [1]. The published description of that check is truncated, so its rejection criteria and any remedies are not stated [1].

AWS also states that cloned voices are automatically encrypted and access-controlled so that only the customer can access them, and requires consent confirmation each time a voice is created [1].

Known, claimed, inferred

Building a cloned voice is free for Amazon Polly customers, while cloned-voice usage is charged on a pay-as-you-go basis [1]. Actual prices, billing units and cost estimates are not provided.

A reasonable hypothesis is that production adoption will depend on governance and voice-quality review as much as on the synthesis API. Every voice requires a consenting speaker, a creation-time quality check and language-specific approval [1].

AWS’s application examples, including consistent brand narration across channels, are proposed use cases rather than documented customer outcomes [1]. The demonstrations in the announcement likewise do not provide independent evidence of voice similarity, naturalness or accuracy.

Limitations & uncertainties

The available evidence comes from a single official AWS announcement and contains no independent third-party test report or customer testimonials [1]. The claimed naturalness and accuracy therefore remain unverified outside AWS’s own description [1].

Exact eligibility criteria for the gated release, how access requests are evaluated and expected approval times are not stated [1]. It is also unclear whether the stated 5–10 minute creation time applies to all seven languages and all supported Regions [1].

Recording specifications, including file formats, sampling rate, maximum duration, environmental conditions and minimum usable audio length, are unspecified [1]. Voice-count limits, treatment of previously approved access requests, and whether customers can delete cloned voices and associated recordings are also unknown [1].

Evidence & sources

The key statements of this analysis, labeled by type, with the sources they rely on. Trendoviz does not run product tests.

  • Primary-source or vendor statementAmazon Polly voice cloning is generally available as a gated release, with access enabled for select customers and available to others by contacting an AWS account team or applying.[1]
  • Primary-source or vendor statementThe service supports seven languages, but each cloned voice speaks only the language in which it was recorded.[1]
  • Primary-source or vendor statementA `CreateVoice` request takes 1–10 recordings, their language and a consent confirmation, and returns a voice ID before the voice is built asynchronously.[1]
  • Primary-source or vendor statementAWS says voice creation typically takes 5–10 minutes, during which Polly checks the resulting voice for speaker similarity.[1]
  • Primary-source or vendor statementAWS says cloned voices use the Polly generative engine and deliver naturalness and accuracy similar to its other generative voices.[1]
  • Primary-source or vendor statementBuilding a cloned voice is free for Amazon Polly customers, while cloned-voice usage is billed on a pay-as-you-go basis.[1]
  • Primary-source or vendor statementThe feature is available in all AWS Regions supported by the Amazon Polly generative engine, including AWS GovCloud (US).[1]

Sources

  1. 1

Updates & corrections

  1. First published · approved by an editor

Found an error? Report it.