Amazon Polly
Text-to-speech with lifelike neural and generative voices.
❓ What is it?
A text-to-speech service that renders text into natural audio across dozens of languages and voices — standard, neural, and generative engines — with SSML control over pronunciation, pace, and emphasis, plus lexicons for domain terms.
💡 Why does it exist?
Audio interfaces (IVR, accessibility, content narration) need consistent, scalable voice. Recording humans is slow and unchangeable; Polly synthesises speech on demand and lets content stay editable text.
⏱️ When should you use it?
Use it for IVR prompts, article/audiobook narration, e-learning voiceovers, and accessibility features; pick neural or generative voices when naturalness matters more than cost.
🗺️ Where does it fit?
At the presentation edge: applications call SynthesizeSpeech and play or cache the audio stream (commonly to S3/CloudFront); Amazon Connect uses it for dynamic contact-centre prompts.
🔌 How do you integrate it?
Call SynthesizeSpeech with text or SSML and a voice ID; store MP3/OGG output; use speech marks for lip-sync or highlighting, and lexicons to pin pronunciations of product names.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- SSML and lexicons are the answers for pronunciation-control scenarios.
- Polly is the voice OUTPUT half of a bot: Lex/LLM produces the words, Polly speaks them.