Amazon Polly introduces a significant upgrade to its Generative engine
Amazon Polly upgrades its generative engine with 57 voices, new Hindi/Finnish locales, better prosody and accuracy, and no code or pricing changes.
Key Takeaways
- Amazon Polly’s Generative engine now offers 14 new voices across 9 locales, expanding the catalog to 57 voices in 11 languages and 25 locales.
- Hindi and Finnish are added to the generative voice catalog for the first time, while the upgrade also extends availability to AWS GovCloud (US-West).
- The new voices provide more expressive, context-aware speech and better accuracy for details such as tracking numbers, dates, currency, and addresses.
- The upgrade is a drop-in change: existing applications using generative voices keep the same voice names, API calls, and pricing, with no code edits required.
- The improvements are especially relevant for contact centers, LLM-based voice agents, media localization, e-learning, accessibility, and interactive experiences.
Amazon has upgraded the Generative engine behind Amazon Polly, its fully managed AWS text-to-speech service, adding 14 new voices across 9 locales and bringing Hindi and Finnish to the generative catalog for the first time. The upgrade is generally available in production now, and for anyone already using generative voices it arrives as a drop-in change: voice names and API calls stay the same, so existing applications keep working without code edits.
The headline number for the expanded catalog is 57 generative voices spanning 11 languages and 25 locales. Beyond the two net-new languages, the release extends regional reach to AWS GovCloud (US-West), bringing the total to 11 AWS Regions.
What actually changed
The core claim is better output: more expressive and more accurate speech, delivered with low latency, according to Amazon. The improvements cover two areas that matter in production.
The first is naturalness. Amazon says the voices now reflect emotion and prosody based on the input text without manual setup, so the same voice can shift delivery depending on context. Sample audio in the announcement illustrates this with a neutral line ("Your package arrived early."), an uncertain version ("So, um… your package arrived early. I think?") and a surprised one ("Wait, what? Your package arrived early?").
The second is accuracy on entities. The company demonstrates a voice reading back an order confirmation with a tracking number, date and time, currency and address — the kind of detail that decides whether a self-service call resolves or gets escalated to a human agent.
The new voices, by locale
The 14 additions break down as follows: US English (Kenneth, Nate), Australian English (Russell, Curtis), Indian English (Kiara, Arjun), German (Tina), US Spanish (Carlos), Brazilian Portuguese (Leo, Carolina), Korean (Hyunduk), Hindi (Kiara, Arjun) and Finnish (Suvi).
Hindi and Finnish are the first languages to reach general availability on the Generative engine. Amazon's samples include Arjun reading a news bulletin, Kiara opening a story, and Suvi narrating a lighthouse keeper tale in Finnish.
Why the drop-in detail matters
For teams already running Polly, the practical story is that nothing in the integration changes. The upgrade happens in place for applications using generative voices. New applications use the same `SynthesizeSpeech` API and gain access to the larger catalog. The `StartSpeechSynthesisStream` API supports bidirectional streaming — text in, synthesized audio back simultaneously — which suits real-time work like voice agents and game characters.
Pricing stays flat on the pay-as-you-go model: no cost increase, no new SKU, no changes to existing integrations. Cached and replayed static prompts carry no additional cost.
There are limits worth noting. Standard and neural voices are unaffected by the upgrade. Speech marks remain unavailable on generative voices, unchanged from before — applications that depend on them should stay on standard or neural voices. SSML tags are still supported on the generative engine.
Where the voices fit
Amazon positions the engine for contact centers, LLM-based voice agents, media and localization, e-learning, accessibility and interactive experiences. In contact centers, accurate pronunciation of tracking numbers and confirmation codes supports self-service IVR. In cascaded voice agents, Polly converts LLM response text into speech streamed over WebSockets or telephony channels, with low-latency playback and high scalability as the stated benefits. For media and dubbing, the expanded locale coverage means one pipeline can localize content into more markets — including, for the first time, generative Hindi and Finnish.
Sanoma Media Finland, which publishes over 10,000 articles per month using Polly's Finnish neural voice, tested the new generative voice. Justus Reinikainen, Product Manager at Sanoma Media Finland, said in the announcement that Suvi's generative voice is "a clear improvement over the previous neural voice — it sounds noticeably more natural and makes for a more enjoyable experience for the listener."
Getting started and availability
Developers can access the voices through the AWS Management Console, AWS Command Line Interface (AWS CLI) or the AWS SDKs. Prerequisites are an active AWS account with Polly access, the AWS CLI installed and configured, Python 3.10+, Boto3 installed and IAM permissions for `polly:SynthesizeSpeech`.
A minimal Boto3 call sets `Engine="generative"` and picks a voice — Kenneth, for example — with MP3 output written to a local file. The equivalent AWS CLI command uses `aws polly synthesize-speech` with the same engine and voice flags. To confirm a voice exists in a given Region at deployment time, call `DescribeVoices`.
The 14 new voices are available in 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 AWS GovCloud (US-West). Contact-center teams get the improvements automatically in Amazon Connect and Genesys flows.
For developers, the takeaway is simple: if you already call Polly with the generative engine, you get better voices without touching code. If you have been waiting on Hindi or Finnish, the wait is over — and checking `DescribeVoices` at runtime is the safe way to confirm availability wherever your application runs.

Comments (0)