Amazon Polly Generative engine: 14 new voices, Hindi and Finnish GA

Amazon has upgraded its Amazon Polly Generative engine with improved speech quality, more voice options, and support for two new languages. The company says the upgrade delivers more expressive and accurate speech output with low latency, now generally available in production.

The headline addition is 14 new voices spanning 9 locales. This expands the overall catalog to 57 generative voices across 11 languages and 25 locales. Two of those locales represent first-time support for Hindi and Finnish in the Generative engine, each with dedicated voice options. Hindi gets voices named Arjun and Kiara. Finnish gets Suvi. The upgrade also adds new voices to existing locales including US English, Australian English, Indian English, German, US Spanish, Brazilian Portuguese, and Korean.

The upgrade matters most to teams building contact centers, voice agents, media localization pipelines, and accessibility features. For existing users, there is no friction. According to the announcement, this is a drop-in upgrade where voice names and API calls stay the same, so production applications continue working without modification. New applications can access the expanded voice catalog through the same SynthesizeSpeech API.

What changed

AWS announced the Amazon Polly Generative engine upgrade, now generally available in production. The company says the upgrade delivers more natural, expressive, and accurate speech output alongside improvements in entity pronunciation and low-latency generation. The upgrade adds 14 new voices across 9 locales and marks the general availability of Hindi and Finnish for the first time on the Generative engine. Region availability expands to include AWS GovCloud (US-West), bringing total supported regions to 11.

Sample audio clips demonstrate the improvements in the announcement. One example features a voice named Kenneth reading an order confirmation with accurate pronunciation of tracking numbers, dates, times, currency, and addresses. Another shows how the voices can reflect different emotional contexts based on text input, such as neutral, uncertain, or surprised tones, without manual configuration.

The voices use Amazon Polly’s standard API and work with existing AWS patterns including Identity and Access Management controls, the Bi-Di Streaming API for real-time applications, and Speech Synthesis Markup Language support for fine-grained control. According to the announcement, the upgrade comes at the same price with no cost increase, no new billing SKU, and no changes to existing integrations.

Why it matters

The upgrade affects teams who rely on text-to-speech for customer-facing workflows. Contact center operators gain more accurate speech for interactive voice response systems, which reduces escalation to human agents when customers confirm details like addresses and order numbers. Voice agent builders get more expressive speech output that can stream in near real time over WebSockets or telephony, improving conversational quality.

For media and localization teams, the new Hindi and Finnish support means a single content pipeline can now reach more markets without separate voice talent or studio time. Publishers and e-learning platforms can update narrated content instantly when source text changes. Accessibility teams gain more natural-sounding options to serve a wider audience.

The drop-in nature of the upgrade removes a common adoption friction. Teams using older generative voices do not need to rebuild integrations or test new API parameters. The upgrade applies automatically to existing implementations.

The expansion to AWS GovCloud opens the service to government agencies and regulated enterprises with strict data residency requirements. Sanoma Media Finland, a news publisher, reported in the announcement that the new Finnish voice Suvi sounds noticeably more natural than the previous neural voice and provides a more enjoyable listening experience for readers consuming audio versions of news articles.

What to test

Before adopting the new voices in production, validate these vendor claims:

Test entity pronunciation accuracy with real data from your domain. The source shows order confirmations as an example, but your contact center may use industry-specific terms, product names, or abbreviations that need verification.

Evaluate latency in your network and infrastructure. The source says the upgrade supports low-latency streaming for real-time applications, but measure end-to-end performance from API call to audio playback in your environment.

Compare emotional prosody against your specific use cases. The source demonstrates neutral, uncertain, and surprised tones, but confirm the voices match the tone your applications require.

Test multilingual switching if you plan to use Hindi, Finnish, and other new locales in the same application. Confirm voice quality is consistent across languages and that your text preprocessing handles language-specific characters and punctuation correctly.

Validate SSML fine-grained control with your speech patterns. The source mentions SSML support, but test whether your existing markup rules apply to the new voices or need adjustment.

Check compatibility with your existing cache and replay workflows. The source says static prompts can be cached with no additional cost, but confirm your application’s caching layer works as expected.

Verify real-time streaming performance using the Bi-Di Streaming API if your application converts language model responses into speech for voice agents or other conversational interfaces.

Common questions

What is Amazon Polly?

Amazon Polly is a fully managed AWS service that converts text into natural-sounding speech. According to the announcement, it serves contact centers, large language model-based agentic applications, media and entertainment, e-learning, accessibility, and other applications requiring reliable voice output at scale.

How does the Amazon Polly Generative engine work?

The source explains that the Generative engine is Amazon Polly’s largest text-to-speech model to date. It uses a transformer model to convert raw text into speech codes, then generates audio from those codes. The engine now supports 57 voices across 11 languages and 25 locales.

What is the difference between Amazon Polly and other voice engines?

Amazon Polly has four voice engines, according to the source: Generative, Long-form, Neural, and Standard. The upgraded Generative engine is designed for more natural and expressive speech with low latency, making it suitable for real-time applications like voice agents and contact center interactions.

The conclusion

The Amazon Polly Generative engine upgrade expands voice options and language coverage without breaking existing implementations. The addition of Hindi and Finnish support opens the service to larger addressable markets in South Asia and Northern Europe. Teams building voice agents and contact center systems should test the new voices against their production data to confirm entity accuracy and emotional prosody match their requirements.

The no-cost upgrade and backward compatibility reduce risk for existing users. The expansion to AWS GovCloud signals AWS’s intent to serve regulated industries and government agencies.

Watch for adoption patterns in media localization and voice agent frameworks. If the new voices deliver the accuracy and expressiveness the source claims, multilingual content teams may shift from studio recording or competing services to Amazon Polly for rapid, cost-effective localization at scale.

Compare the new generative voices against competing services like ElevenLabs and Murf AI to confirm whether Amazon Polly’s pricing and quality match your workflow. For teams building voice agents, also evaluate real-time performance using the Bi-Di Streaming API against your latency requirements.

Related reading

• Voxtral 3 voice agent SDK: 50+ languages and emotional speech
• Granite Speech 5.0 2B ranks #2 on Open ASR Leaderboard