Amazon Polly Generative engine upgrade adds 14 voices, first Hindi and Finnish support, at unchanged pricing

Amazon Web Services has upgraded the Amazon Polly Generative engine, adding 14 new voices across 9 locales and bringing Generative voices to Hindi and Finnish for the first time. The upgrade is generally available in production and applies in place to existing generative voices, with no code changes required.

What changed

According to AWS, the upgraded engine produces "more expressive, more accurate speech output with low latency." AWS has not published benchmark scores, parameter counts, or latency figures to support these claims. It offers audio samples instead.

The announced improvements fall into three areas:

• Expressiveness: AWS says voices now reflect emotion and prosody based on the text input, without manual setup. Its examples include neutral, uncertain, and surprised deliveries of the same sentence.
• Entity accuracy: The upgrade targets numbers, currency, addresses, dates, and alphanumeric codes. AWS's demo has the voice "Kenneth" read back a tracking number, a street address, a dollar total, and a confirmation code.
• Language coverage: Hindi (Arjun, Kiara) and Finnish (Suvi) are now generally available on the Generative engine.

Catalog and regions

The Generative engine now offers 57 voices across 11 languages and 25 locales. The 14 new voices are:

• 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
• Finnish: Suvi

The engine is also now available in AWS GovCloud (US-West), bringing the total to 11 AWS Regions.

API and pricing

Voice names and API calls are unchanged. Existing applications using generative voices pick up the new quality automatically. New applications use the same SynthesizeSpeech API. For real-time use cases such as voice agents and game characters, the bidirectional streaming API StartSpeechSynthesisStream lets developers stream text in and receive audio back simultaneously. SSML support remains available for fine-grained control.

AWS says pricing is unchanged: no cost increase, no new SKU. Specific per-character rates were not included in the announcement, so they are not restated here; AWS directs readers to the Amazon Polly pricing page. AWS also says synthesized static prompts can be cached and replayed at no additional cost.

AWS did not disclose a context window, parameter count, training data cutoff, or version identifier for the upgraded engine.

Customer evidence

AWS cites Sanoma Media Finland, which says it publishes over 10,000 articles per month as audio using Polly's Finnish neural voice. Based on its own testing, Sanoma describes Suvi's new generative voice as "a clear improvement" that sounds "noticeably more natural." This is a single customer testimonial, not an independent evaluation.

What this means

This is a quality and coverage update to a managed service, not a new model release. AWS has not exposed a new version string, so developers cannot pin to the old or new behavior. Because the upgrade is silent and in place, teams with regression tests on audio output, such as IVR flows with fixed prompts, should re-evaluate their outputs.

The emphasis on entity accuracy and streaming targets contact centers and cascaded voice-agent stacks, where an LLM generates text and a separate TTS engine speaks it. In that architecture, misread order numbers and addresses are the failure mode that sends callers to human agents. Holding the price flat while improving quality also puts pressure on rival TTS providers. The addition of GovCloud (US-West) widens the engine's reach to regulated US government workloads.

The main gap is evidence. Without published naturalness scores, word-error or entity-accuracy rates, or latency numbers, the improvements remain AWS's claims until developers test them against their own content.