Amazon has upgraded Polly’s generative text-to-speech engine with 14 voices across nine locales, bringing its catalog to 57 generative voices in 11 languages and 25 locales. Hindi and Finnish generative speech are generally available for the first time.
The new engine is a drop-in replacement for existing generative voices: voice names, API calls and pricing remain unchanged. Existing applications receive the upgrade in place, while new projects continue using the SynthesizeSpeech API. A bidirectional streaming API can accept text and return audio at the same time for live agents and other low-latency uses.
AWS says the model improves expression and pronunciation of difficult entities such as addresses, currencies and tracking numbers. The new voices cover US, Australian and Indian English, German, Spanish, Portuguese, Korean, Hindi and Finnish. Availability also expands to AWS GovCloud in the US West region, making 11 supported regions in total.
The release is generally available for production, but generative voices still do not provide speech marks, which applications use to synchronize words and animation. Standard and neural Polly voices retain that feature, so teams that depend on precise timing cannot switch engines without checking that limitation.