Amazon Polly's Generative engine has received an upgrade that AWS announced on its Machine Learning Blog. It is generally available in production now.
What changed
AWS says the engine produces more expressive and more accurate speech at low latency. The voices are meant to adjust emotion and prosody to the text itself, without manual setup. The examples show the same sentence, "your package arrived early", delivered in neutral, uncertain and surprised tones.
Accuracy with entities is a stated focus. The sample audio has a voice reading back an order confirmation with a tracking number, date and time, currency amount, address and alphanumeric confirmation code.
Voices, languages and Regions
Fourteen new voices span nine locales: US, Australian and Indian English, German, US Spanish, Brazilian Portuguese, Korean, Hindi and Finnish. Hindi (Arjun and Kiara) and Finnish (Suvi) are generally available on the Generative engine for the first time. The catalog now totals 57 generative voices across 11 languages and 25 locales. AWS GovCloud (US-West) joins the supported Regions, bringing the total to 11.
What stays the same
This is a drop-in upgrade. Voice names and API calls are unchanged, and applications already using generative voices are upgraded in place. AWS says the price is unchanged, with no new SKU, and that static prompts can be cached and replayed at no additional cost. The Bi-Di Streaming API, StartSpeechSynthesisStream, lets an application send text and receive audio at the same time, which AWS positions for real-time voice agents.
AWS also says the improvements reach Amazon Connect and Genesys flows automatically.
Limits to know
Standard and neural voices are unaffected by the upgrade. Speech marks are not available on generative voices, so applications that need them have to stay on standard or neural voices. SSML tags continue to be supported. AWS recommends calling DescribeVoices at deployment time to confirm a chosen voice is available in the Region where the application runs.
A customer data point
Sanoma Media Finland, which publishes over 10,000 articles a month as audio using Polly's Finnish neural voice, told AWS that Suvi's new generative voice is a clear improvement and sounds noticeably more natural. That is a vendor-published testimonial based on the customer's own testing.
Top comments (0)