Amazon: Amazon Polly Generative

Amazon Polly's largest model — a billion-parameter transformer that turns text into speech codes, decoded by a streamable convolutional decoder — producing the most human-like, emotionally engaged conversational voices: 43 voices across 23 documented locales (24 language codes observed via DescribeVoices, including Hindi through the bilingual voice Kajal). No speech marks, no newscaster style, and a reduced SSML subset.