AWS Polly cost: per character, standard vs neural voices

Quick answer

Polly bills per character synthesized to speech: standard voices are cheaper (about $4 per million characters) and neural voices cost several times more (about $16 per million), with a free tier. Cost scales with text synthesized, so caching generated audio, using standard voices where quality allows, and not re-synthesizing unchanged text are the levers.

Amazon Polly converts text to lifelike speech, billed per character of input. The main cost decision is standard versus neural voices, standard is cheaper, neural sounds more natural but costs several times more, and the main saving is not re-synthesizing the same text.

Standard vs neural

Voice type	Rate (approximate)
Standard voices	~$4 / million characters
Neural voices	~$16 / million characters
Free tier	Monthly character allowance

Standard voices cost about $4 per million characters; neural voices, which sound more natural, cost about $16 per million, roughly four times more. A free tier covers initial usage. So choosing neural where standard would be acceptable multiplies the per-character cost.

Cache generated audio

The biggest saving is caching. Synthesized audio for a given text and voice is deterministic, so storing the generated audio file keyed by the text and voice means repeated content, notifications, common phrases, static articles, is synthesized once and served from storage thereafter. For content that does not change, this eliminates repeated synthesis cost.

Controlling Polly cost

Cache generated audio and serve it from storage rather than re-synthesizing, use standard voices where their quality is acceptable and reserve neural for content where naturalness matters, and synthesize only text that will be heard. As with other per-character AI services, the levers are voice choice and caching, mirroring Translate.

FAQ

How is AWS Polly priced?

Per character of text synthesized to speech, with standard voices costing about $4 per million characters and neural voices about $16 per million (roughly four times more), plus a monthly free tier. The voice type and text volume together determine cost.

Are Polly neural voices worth the extra cost?

Where naturalness matters, such as customer-facing or long-form content, the more lifelike neural voices can justify the roughly four-times-higher per-character cost. For internal, transactional, or short notifications where standard voice quality is acceptable, standard voices are far cheaper. Match the voice to the use case.

How do I reduce Polly cost?

Cache generated audio and serve it from storage rather than re-synthesizing the same text, use standard voices where their quality suffices and reserve neural for content where naturalness matters, and synthesize only text that will actually be heard. Voice choice and caching are the main levers.

Does caching help with Polly cost?

Yes, significantly. Synthesized audio for a given text and voice is deterministic, so storing the audio file keyed by text and voice means repeated content, notifications, common phrases, static articles, is synthesized once and served from storage thereafter, eliminating repeated synthesis cost for unchanging content.

What is the difference between Polly standard and neural voices?

Standard voices use concatenative synthesis and cost less (about $4 per million characters); neural voices use a machine-learning model for more natural, human-like speech and cost more (about $16 per million). Neural sounds better but costs roughly four times as much, so choose by quality need.

Does C3X estimate Polly cost?

Polly cost is driven by characters synthesized and voice type, which are usage inputs. C3X prices the surrounding infrastructure, and you model your synthesis volume and voice choice to estimate the per-character text-to-speech charges.

What to do next

Estimate the infrastructure around your speech pipeline before you build it. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.