Best Free TTS APIs in 2026 (Compared & Tested)

The quick comparison

* edge-tts uses an unofficial Microsoft endpoint; limits are undocumented and availability can change. Figures reflect each provider's published terms as of August 2026 - always re-check before committing.

1. eidosSpeech - genuinely free, no card

Full disclosure: this is our product. eidosSpeech wraps Microsoft Edge's neural voices behind a simple REST API and a browser studio. Registration takes an email - no credit card - and the free tier is the whole product, not a trial: 30 requests/day, 1,000 characters per API call (2,000 in the web UI), 1,200+ voices across 75+ languages, and the endpoint returns MP3 directly, so there's no audio-format plumbing on your side.

curl -X POST https://eidosspeech.xyz/api/v1/tts \
  -H "X-API-Key: esk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"text":"Hello world","voice":"en-US-JennyNeural"}' \
  --output hello.mp3

The honest downsides: 30 requests/day won't cover bulk pipelines, and voice realism is Edge-neural quality - excellent, but a step below ElevenLabs' best models. If you outgrow the limit, the options are registering a second workflow or moving to a cloud free tier below.

2. edge-tts - unlimited, but unofficial

The Python edge-tts library speaks the same Microsoft Edge neural voices with no account at all, and it's the engine behind several "free unlimited" tools. The catch is the asterisk: it targets an endpoint Microsoft never publicly documented as an API. It can throttle, break, or change without notice - fine for personal scripts you maintain, risky for anything customer-facing. eidosSpeech uses the same voice family but fronts it with a stable, documented API and rate-limit headers.

3. Kokoro-82M - open weights, self-hosted quality

Kokoro is an 82-million-parameter open-weight model (Apache 2.0) that punches far above its size - natural English, plus a growing set of languages. It's free forever because it runs on your hardware. The cost shifts to infrastructure: you host the model, you manage scaling, and the language list is shorter than the cloud catalogs. Best pick if data privacy or unlimited volume matters more than zero-ops.

4. Piper - lightweight, runs anywhere

Piper (MIT) is a fast, lightweight TTS that runs on a Raspberry Pi. Dozens of community voices across many languages. Quality is clearly more synthetic than neural cloud voices, but for kiosk announcements, home automation, or embedded projects where latency and hardware cost matter, it's unbeatable.

5. ElevenLabs - best realism, tightest free cap

ElevenLabs' voices are the most natural on this list, and the free tier needs no card. But it caps around 10 minutes of audio per month and requires attribution to ElevenLabs when publishing. For a short hero voiceover it's fantastic; for daily content volume the cap runs out in days.

6. Google Cloud TTS & AWS Polly - free tiers, not free products

Both hyperscalers give real monthly allowances (about 1M WaveNet characters on Google, 5M characters on AWS - the latter only for the first 12 months). Both require an account with billing enabled, so "free" here means "free until you cross a threshold or the trial window closes." If you're already on GCP or AWS and volume is high, they're excellent engineering choices; if you just want a key that works without a card, look at the options above.

Which should you pick?

• Daily videos, podcasts, indie apps - no card: eidosSpeech.
• Personal automation you'll babysit: edge-tts.
• Unlimited volume or private data: Kokoro-82M (quality) or Piper (tiny hardware).
• One-off ultra-realistic clips: ElevenLabs free tier.
• Enterprise pipelines already on a cloud: Google Cloud TTS or AWS Polly.

Frequently asked questions