Amazon Polly is the text-to-speech service in AWS. It turns text into spoken audio in dozens of voices across more than 25 languages and is reached through the service API and language-specific SDKs, the AWS Management Console, or the AWS CLI. Audio can be streamed back in near real time, which is how voice-agent frameworks use it as a speech output stage.
Polly offers four voice engines: standard, neural (NTTS), long-form for extended reading, and generative for the most conversational voices. Speech is shaped with SSML markup, custom pronunciation lexicons, and speech marks that report when each sentence, word, and phoneme is spoken.
Features
- Voice engines: standard, neural, long-form (voices such as Danielle, Gregory, Ruth, Patrick, Alba, and Raúl), and generative
- Languages and voices: 25+ languages including Arabic, Cantonese, Mandarin, Hindi, Japanese, Korean, and several English, Spanish, French, and Portuguese variants, with male, female, and regional options
- SSML: the W3C markup for controlling phrasing, emphasis, pauses, and intonation
- Speech marks: metadata stream with timing for sentences, words, and visemes, used for lip-sync and highlighting
- Custom lexicons: XML lexicons that override pronunciation of company names, acronyms, and foreign words
- Output formats: MP3, Ogg Vorbis, and raw PCM at 8, 16, 22.05, and 24 kHz
- Streaming: audio is returned as a stream for near real-time playback
- Brand Voice: a custom neural voice built for exclusive use by one organization
- Access: API and SDKs, console, and CLI
