What it is
Amazon Polly is AWS's text-to-speech service. It turns text into natural-sounding speech in many voices and languages and returns audio such as MP3. It offers several engines: standard, neural, long-form and generative, with the generative engine giving the most human-sounding delivery. Applications call it through the AWS API and can play or save the audio it returns.
Speech can be controlled with SSML for pauses, emphasis and pronunciation, and Polly can return timing marks so text on screen can follow the voice. Pricing is per character converted, so short sentences cost very little, and nothing is billed when no one is listening. There is no model to host and no capacity to reserve.
How I use it
- On the ask page, tap the microphone and ask out loud. The browser turns speech into the question.
- The answer comes back in an Amazon Polly generative voice, sentence by sentence while it is still being written.
- Speech requests go through the same Lambda function, CloudFront and daily caps as the rest of the site.
Polly is the right tool when an application needs to talk and you want to pay per sentence rather than run a speech model. I use it so the agent can answer out loud, and I send it a sentence at a time so speech starts before the answer is finished.
Questions
Can an AI assistant answer out loud?
Yes. My agent's answers are read by an Amazon Polly generative voice one sentence at a time, so it starts speaking before the answer is complete.
Is text-to-speech expensive?
Polly charges per character, so a spoken answer costs a fraction of a cent, and nothing is spent when no one asks.