AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Amazon Polly delivers a major upgrade to its Generative engine

Collected Oct 22, 2026

AWS has released a production upgrade to the Generative engine behind Amazon Polly, its managed text-to-speech service. The company says the new engine produces more expressive and more accurate speech at low latency, and that the upgrade is generally available in all supported Regions. Because it is a drop-in change, existing generative voice identifiers and API calls still work, and the new voice renderings simply replace the old ones in place.

The headline addition is 14 new voices spanning 9 locales: Kenneth and Nate for US English, Russell and Curtis for Australian English, Kiara and Arjun for Indian English, Tina for German, Carlos for US Spanish, Leo and Carolina for Brazilian Portuguese, Hyunduk for Korean, Kiara and Arjun for Hindi, and Suvi for Finnish. Hindi and Finnish are generally available on the Generative engine for the first time. With these additions, the generative catalogue reaches 57 voices across 11 languages and 25 locales. The launch also extends to AWS GovCloud (US-West), bringing the total to 11 Regions: US East (N. Virginia), US West (Oregon), Europe (Frankfurt), Europe (Zurich), Europe (London), Asia Pacific (Sydney), Asia Pacific (Seoul), Asia Pacific (Singapore), Asia Pacific (Tokyo), Canada (Central) and GovCloud US-West.

Much of the improvement is aimed at the failure modes that make synthetic speech obvious in customer-facing systems. AWS highlights accurate reading of numbers, currency amounts, addresses, dates and alphanumeric codes, and offers a sample of the Kenneth voice reciting an order confirmation complete with order number, street address, ZIP code, delivery window, dollar total and confirmation code. The engine also claims to infer delivery style from the text itself: the same sentence can come out neutral, hesitant or surprised depending on how it is written, without manual configuration. This suggests the model is picking up pragmatic cues from punctuation, discourse markers and phrasing rather than requiring explicit style tags.

On the developer side, nothing changes structurally. Speech is still requested through SynthesizeSpeech with Engine set to "generative", plus a VoiceId and output format. Access control runs through IAM with the polly:SynthesizeSpeech permission, and the service is reachable from the AWS Management Console, the AWS CLI or the AWS SDKs. A minimal Python setup needs an active AWS account, Python 3.10 or later and boto3. Real-time applications can use the bidirectional streaming API StartSpeechSynthesisStream, which accepts text and returns audio concurrently, a pattern suited to voice agents and game characters where waiting for a full synthesis before playback would add noticeable delay. For deployments across Regions, AWS recommends calling DescribeVoices at runtime to confirm a chosen voice exists where the application runs.

A few constraints carry over unchanged. Standard and neural voices are not affected by the upgrade and keep their existing behaviour. Speech marks are still unavailable on generative voices; applications that depend on them, such as word-level highlighting for read-along features, need to stay on standard or neural voices. SSML tags remain supported on the generative engine for finer control over pronunciation and pacing, though the tag set is narrower than what standard voices accept.

Pricing stays flat. AWS describes the commercial terms as pay-as-you-go with improved quality and accuracy at the same price, no new SKU and no change to existing integrations, with cached and replayed static prompts incurring no additional charge; the pricing page carries the current per-character rates. The upgrade also propagates automatically into Amazon Connect and Genesys contact-centre flows, so teams running those stacks receive the new audio without reconfiguration.

Sanoma Media Finland, which converts real-time news into audio and publishes more than 10,000 articles per month using Polly's Finnish neural voice, tested the new Suvi generative voice. Product manager Justus Reinikainen called it a clear improvement over the previous neural voice, noticeably more natural and more enjoyable for listeners. That is a single customer report rather than a measured benchmark, so treat it as directional.

This fits a broader shift in the TTS market toward models that infer prosody and emotion from raw text instead of relying on hand-tuned markup. The practical appeal for voice agents is a cascaded pipeline in which a large language model produces text and Polly speaks it, streamed back over WebSockets or telephony. A likely trade-off is latency: generative synthesis is richer than neural synthesis, and while AWS stresses low latency, the bidirectional streaming path is the one designed to hide it, so teams building interactive agents should test it in their own channels rather than assume batch synthesis timings carry over. The missing speech marks are another constraint worth checking before migrating read-along or karaoke-style features.

Why it matters: Teams already using Polly generative voices get better pronunciation of transactional details and new languages without touching code, which mainly benefits contact centres, voice agents and media localisation pipelines. Product teams that held back on generative voices because Hindi or Finnish were missing can now evaluate them on the same SynthesizeSpeech call. The main checks before deploying are per-Region voice availability via DescribeVoices and whether an application depends on speech marks, which remain exclusive to standard and neural voices.

Read at AWS Machine Learning Blog

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The Amazon Polly Generative engine gets a major upgrade: more expressive, more accurate speech with low latency, plus 14 new voices across 9 locales and general availability of Hindi and Finnish. This drop-in upgrade requires no code changes and is now generally available in production.