AWS announced on October 22 an upgrade to the Amazon Polly Generative engine, which it says produces more expressive and more accurate speech at low latency. The update adds 14 new voices across 9 locales, including the first generally available generative voices for Hindi, voiced by Arjun and Kiara, and for Finnish, voiced by Suvi. AWS describes it as a drop-in change: voice names and API calls are unchanged, so existing applications keep working. Two areas are highlighted as improved, more natural prosody and correct reading of numbers, currency, addresses and dates, and the post publishes sample audio for both, including an order-confirmation clip containing a tracking number, street address, ZIP code, timestamp and total. The engine targets contact centres, voice agents built on large language models, media, e-learning and accessibility workloads.
Our takeDrop-in compatibility is the substantive part: a speech upgrade that renames voices or changes parameters forces every integration to be re-tested, and keeping both stable removes that work. The quality claims stay vendor-side, though. The samples are hand-picked clips, and reading a tracking number, ZIP code or currency correctly is a narrow, testable behaviour that a small regression set of numeric and address strings would settle faster than any listening demo. For voice agents, latency under streaming matters more than prosody.