Free Text-to-Speech APIs in 2026

A text-to-speech (TTS) API turns a string into an audio file or stream. Six verified offers in the catalog let you do that for free, in two shapes: a monthly allowance that refills (ElevenLabs, Cartesia, Amazon Polly) and a one-time credit you spend down (Deepgram, Hume AI, Sarvam AI). Every number below was checked on the vendor's own pricing or docs page on 2026-10-19; where the vendor page did not state a figure, the guide leaves it out.

Compare the free TTS APIs

The "what that buys" column for Deepgram and Sarvam is our arithmetic from the published pay-as-you-go rate; the vendors state the credit and the rate, not the character count.

Which one to pick

• Most free speech every month, for good: Amazon Polly's Standard voices, 5M characters a month with no stated end date. The more natural Neural, Long-Form and Generative voices are free only for the first 12 months. Accounts created since 15 July 2025 also choose a free plan (6 months, up to $200 in Free Tier credits) or a paid plan; the free plan ends with the account suspended unless you upgrade.
• Most natural voices on a monthly plan: ElevenLabs or Cartesia. Both refill monthly with no card, and both keep the commercial-use license on paid plans only, so treat the free plan as a prototype budget.
• Biggest one-time budget: Deepgram's $200 credit, which does not expire and also covers speech-to-text, so one key serves a whole voice app.
• Expressive voices and cloning to try: Hume AI's $20 credit; Hume says Octave can clone a voice from about 15 seconds of audio.
• Indian-language apps: Sarvam AI's INR 100 credit, which never expires and also works for its speech-to-text, translation and LLM APIs.

Steps

1. Pick an offer from the table and create an account. On AWS, the free allowance applies to Polly once your account exists; you choose the free or paid plan at sign-up.
2. Create an API key in the provider's console and store it in an environment variable, never in client-side code: anyone holding the key spends your credits.
3. Synthesize one short sentence first and play it back to check the voice, format and sample rate. On ElevenLabs' Free plan audio is capped at 128 kbps, 44.1 kHz.
4. Count characters, not requests. Every provider here meters TTS by text length, so cache audio you will play again (fixed prompts, menus, greetings) instead of synthesizing it on every call.
5. Stay under the concurrency limit: ElevenLabs and Cartesia allow 2 requests at once on their free plans, so queue requests instead of firing them in parallel.

Need speech-to-text too?

Cartesia's free credits and Deepgram's $200 cover both directions. For transcription-only grants, the free speech-to-text API guide compares AssemblyAI, Gladia, Speechmatics, Groq's free Whisper and more, and the free AI voice and audio tools hub lists every audio offer.

When to self-host

When a free allowance runs out or you need audio you can ship commercially without a plan, run an open model locally. The catalog links Piper TTS (GPL-3.0) as the self-hosted stand-in for ElevenLabs, Cartesia, Amazon Polly and Hume: it has no character quota, but no voice cloning or emotional control either.

Limits and the catch

• No commercial license on the monthly free plans. ElevenLabs lists it from Starter and Cartesia from Pro ($5 a month), which is also where Cartesia's instant voice cloning starts; professional voice cloning needs its Startup plan.
• Polly's best voices are a 12-month trial. Only the Standard voices' 5M characters carry no stated time limit.
• One-time credits do not refill. Deepgram's, Hume's and Sarvam's credits are spent once; after Deepgram's, billing continues at pay-as-you-go rates.
• Free tiers change. Offers are re-checked monthly; the verified date on each offer page is the day we last confirmed it.