Amazon Polly generative, long-form, neural, and standard speech synthesis.
6 integrated models · Official documentation ↗
Amazon's December 2025 speech-to-speech model on Bedrock for live voice conversations over a bidirectional stream, with barge-in, adjustable pause sensitivity and async tool calls in seven languages; 1M-token context. Superseded by Nova 2.5 Sonic on 2026-10-05.
Amazon Polly's top engine, a billion-parameter transformer generally available since May 2024, now with 43 voices across 23 language variants and bidirectional streaming. It does not return speech marks, which the other Polly engines offer.
Polly engine for long passages such as articles and training material, launched November 2023; six voices (four US English, two Castilian Spanish), us-east-1 only, with speech marks. No voices have been added since November 2024.
Polly's neural TTS engine (July 2019), a sequence-to-sequence model with a neural vocoder and the engine AWS still expands most: dozens of voices in 36 languages and variants, 19 regions, and a Newscaster style for four voices.
AWS's managed speech-to-text service, generally available since April 2018, for batch files and real-time streams. Since late 2023 it runs on a multi-billion-parameter speech foundation model covering 100+ languages, with diarization of up to 30 speakers.
Polly's original concatenative engine from the service's November 2016 launch, now with 60 voices in 29 languages and variants across 24 regions. The cheapest Polly engine, with speech marks; neural and generative engines sound more natural.
