Amazon Polly vs Azure Speech in Foundry Tools 2026

Published September 2, 2026 · This is a documentation-based comparison. Every figure below was read from the two vendors’ own sources on September 1, 2026 and carries that check date. No listening test has been run, so nothing here compares how the two providers sound.

The short answer

On this page

Winners by category

Every winner below is a documentation-based judgment from published prices, terms and feature documentation. None is a test result, and the categories that require listening have no winner.

Price, and where Azure’s comes from

On the standard tiers Polly is 3.75 times cheaper: $4.00 against $15.00 per million characters. At Polly’s Neural tier ($16.00) the two are close, and above that they diverge again — Azure’s Neural HD is $22.00 while Polly’s Long-Form reaches $100.00.

But the more important fact is where Azure’s number comes from. Microsoft’s pricing page publishes labels, billing-unit footnotes and free-tier allowances as ordinary text — and renders every actual price as the literal string “$-”. We counted 142 such cells; the only dollar-and-digits string on the entire page is “$200”, which is free-account credit boilerplate. The page also carries its own caveat: “Prices are estimates only and are not intended as actual price quotes.”

The $15.00 and $22.00 figures therefore come from Microsoft’s own first-party Azure Retail Prices API, queried for the Foundry Tools service in the East US region, and the $15 rate is separately corroborated by a Microsoft Learn cost-estimation document stating in plain prose: “Multiply the result by the unit price of $15 per million characters to estimate the monthly cost.” Two independent Microsoft sources agreeing is why we publish it at all.

What this means for you: the number you take to a budget meeting cannot be sourced to the page your finance team will open. Point them at the Learn document instead. Amazon has no equivalent problem — its rates are text on the pricing page — though note that Polly carries a separate, higher GovCloud rate card at $4.80 and $19.20, with only two engines listed.

Free tiers

Amazon offers ten times more, with an asterisk. Five million characters a month on Standard voices, against Azure’s half a million. But AWS states the duration of that Standard allowance two different ways: the pricing page attaches “for the first 12 months” to its Neural, Long-Form and Generative allowances and not to Standard, while the FAQ describes the whole free tier as twelve-month. Budget on the twelve-month reading.

Azure’s is smaller, recurring and clearly stated: “Text to Speech / (per character billing) / Neural / 0.5 million characters free per month.” Two conditions travel with it. Unused custom models are decommissioned after seven days. And a much larger legacy figure — five million characters a month — still displays on the same page inside a block Microsoft labels as deprecated and available only to existing customers. If you have seen Azure quoted as offering five million free characters, that is where the number comes from, and a new customer cannot have it.

Voice cloning: a closed door and a custom one

Neither provider sells cloning self-serve, and the difference between them is eligibility versus price.

Azure’s is an eligibility gate. Its responsible-AI documentation states: “As a Limited Access feature, access to custom neural voice requires registration. Only customers managed by Microsoft, meaning those who are working directly with Microsoft account teams, are eligible for access.” That covers professional voice fine-tuning, personal voice and custom avatar alike. If you do not have a Microsoft account team, no tier and no budget unlocks it.

Amazon’s is a custom engagement. Brand Voice is described as work you do with the Polly team to build a voice “for the exclusive use of your organization”. No price, sample requirement, turnaround or minimum commitment is published anywhere we read, and the pricing page never mentions it.

Both are defensible abuse controls. Neither is available to an independent developer this week. If self-serve cloning is the requirement, ElevenLabs documents it from $6 a month and Cartesia from $5 (both checked September 1, 2026).

Deployment and breadth

This is Azure’s side of the trade, and it is a real one.

Microsoft documents container images and on-premise deployment for speech synthesis, which almost nothing else in our comparison set offers at any price. If your requirement is that audio never leaves your infrastructure, Polly cannot meet it and Azure can. Microsoft also documents the broadest voice and language catalog of any provider we cover, alongside mature SDK coverage and batch synthesis.

Polly’s answer is regional breadth rather than deployment flexibility: 24 documented regions on its Standard engine, including GovCloud and a China region — though its most expensive engine, Long-Form, is documented in exactly one region.

Documentation reliability

Both vendors contradict themselves, and a buyer should know where.

Polly’s contradictions are about terms. Its free-tier duration has two official answers. Its product page states it “does not retain the content of your text submissions” while Service Terms 50.3 provides for AWS using and storing that content to improve its services. Its own pages give conflicting valid sample rates and even conflicting voice genders.

Azure’s are about scale. Three official pages give three different totals for voices and languages. A single page states three different high-definition voice counts. Individual voice families are given conflicting language counts within one page. Regional availability of HD voices differs between the regions page and the feature documentation. And the product has been renamed twice — it is now Azure Speech in Foundry Tools, formerly Azure AI Speech and before that Azure Cognitive Services Speech — while legacy naming remains live across SDK namespaces, the container registry path, RBAC roles and the Azure resource type.

We could not establish a date for that rename. The documentation carrying the new name is stamped January 30, 2026, but that is page metadata, and a full search of the release notes found no dated rebrand entry.

Voice quality, pronunciation and naturalness

We have not compared how these two providers sound.

Our standardized listening benchmark is designed but not running. Until it does, this page carries no naturalness verdict, no pronunciation comparison and no scores for either provider. Both vendors publish quality claims of their own; those belong to them, and we do not repeat them as findings.

Choose Polly if · Choose Azure if

Choose Amazon Polly if:

• Cost is the constraint. $4.00 against $15.00 per million characters on the standard tiers.
• You need a price you can cite from the vendor’s own pricing page. Azure’s does not show one.
• Your legal review wants ownership in a numbered clause. Service Terms 50.2 provides it.
• You generate the same audio repeatedly — caching and replay are explicitly free.

Choose Azure Speech in Foundry Tools if:

• You need container or on-premise deployment. Polly is cloud only.
• You need the broadest catalog and language coverage, accepting that the exact counts are contested.
• You have a Microsoft account team — which is what unlocks cloning and a real negotiated price.
• You object to inputs improving a vendor’s models by default and cannot rely on an AWS Organizations opt-out being set.

Sources and what we could not verify

Every figure on this page was read from the two vendors’ official sources on September 1, 2026. Full detail is in the individual reviews: Amazon Polly review and Azure Speech in Foundry Tools review. How we verify anything is described on the methodology page.

What we could not verify

Honesty about gaps beats a page that looks complete. Genuinely unresolved as of September 1, 2026:

• How either provider sounds. No listening test has been run.
• Any Azure price on Microsoft’s own pricing page. All 142 price cells render as a placeholder; our figures come from Microsoft’s retail API and a Learn document.
• Whether Amazon’s Standard free tier is perpetual or twelve months. Two official pages disagree.
• Azure’s default position on using inputs for model improvement. Not established from the pages we read; we assert nothing either way.
• Azure’s total voice and language counts. Three official pages give three different figures.
• The date of Azure’s rename. No dated rebrand entry exists in the release notes.
• Brand Voice pricing, sample requirements or turnaround. Not published anywhere.
• Azure custom-voice pricing. Not reachable without the access registration.
• Any latency figure or SLA for Polly’s text-to-speech. Not published on any page we read.

Change log

September 2, 2026 — First publication as a documentation-based comparison. All figures read from official vendor sources on September 1, 2026 and dated accordingly.

Published September 2, 2026 · When our audio benchmark runs, this page gains a head-to-head listening test using the same prompts and documented conditions for both providers, and the naturalness and pronunciation rows gain real winners (methodology). Independence note: this page contains no affiliate links, and no vendor paid for placement or influenced the order — see how we make money and our editorial policy.