Amazon Polly
Amazon Polly is AWS's managed text-to-speech API, GA since 2016, maintained by Amazon Web Services as part of its AI/ML portfolio. Per AWS's own voice table (checked July 2026), it ships 100+ voices across 42 languages/language variants via four engines: Standard ($4/1M chars), Neural ($16/1M), Long-Form ($100/1M), and Generative (~43 voices, $30/1M). In March 2026 AWS added 10 new Generative voices (Tiffany, Brian, Aria, Jasmine, Florian, Lorenzo, Lennart, Sabrina, etc.), 2 new regions, and a Bidirectional Streaming API over HTTP/2 for LLM-driven conversational audio. Free tier: 5M Standard / 1M Neural / 500K Long-Form / 100K Generative characters per month for 12 months. Billing is pay-as-you-go proprietary AWS pricing, no self-hosting. Embedded in AWS Connect IVR, audiobook pipelines, and accessibility tooling across thousands of AWS customers. Current version/status: Generative engine expanded March 2026 (10 new voices + Bidirectional Streaming API); service GA since 2016. License: proprietary SaaS (AWS pay-as-you-go). Pricing: Pay-as-you-go per engine: Standard $4/1M chars, Neural $16/1M, Long-Form $100/1M, Generative $30/1M; free tier (first 12 months) is 5M Standard, 1M Neural, 500K Long-Form, or 100K Generative characters/month. Maintained by Amazon Web Services.
Pick Polly when you're already living in AWS and need TTS wired straight into Connect, Lambda, or S3 pipelines without onboarding a new vendor — Standard voices at $4/1M characters undercut everyone for bulk IVR prompts. Skip it for anything expressive or voice-cloned: ElevenLabs' voices beat Polly's Generative engine on naturalness in most side-by-side tests, and Polly has no instant voice-cloning-from-sample like ElevenLabs or Azure Custom Neural Voice. AWS's own pricing tells the story — Standard/Neural sound robotic enough that they push you toward the $30/1M Generative tier just to sound human, a 7.5x markup over Standard for the same words. Known weakness: AWS's own docs and pricing tiers implicitly admit Standard/Neural voices sound robotic by comparison, steering users to the 7.5x-pricier Generative engine for natural prosody — and even Generative lacks the instant voice-cloning ElevenLabs and Azure Custom Neural Voice offer.