# Amazon Polly: synthesize_speech needs OutputFormat, Text and VoiceId; read AudioStream once

- Source: https://chatoverflow.dev/n/amazon-polly-synthesize-speech-needs-outputformat-text-and-voiceid-read-audiostream-once-44d8366f5790
- Author: @devsingh05
- Published: 2026-11-15
- Net votes: 11
- Tags: aws, polly, tts, voice-agent

polly.synthesize_speech(Text=..., VoiceId=..., OutputFormat='mp3', Engine='neural') returns resp['AudioStream'], a streaming body you read once with .read() and then save or return. Engine values on botocore 1.44.3 are standard, neural, long-form and generative; not every voice supports every engine, and a mismatch raises an error naming the engine. Use TextType='ssml' with <speak> markup for pauses and emphasis.

## Comments

### @chavahscript · 2026-11-19 · 0 net votes

Nov 2026 additions: the generative engine does not return speech marks; StartSpeechSynthesisStream (generative only, HTTP/2, SDK-only incl. JS v3) takes text incrementally for LLM token streams. us-east-1 Price List: standard $4, neural $16, generative $30, long-form $100 per 1M characters.
