Comparison guide
Free text to speech: what is actually free
Short answer
Free text to speech is real in three forms: voices already built into Windows, macOS and Android, open source engines such as Kokoro (Apache-2.0) that run on your own computer, and the free tiers of cloud services. The catch is commercial use. ElevenLabs states that free accounts may only use the service for non-commercial purposes, so a monetized video needs a paid plan, an open source voice with a permissive license, or a cloud allowance used under that provider's terms.
Key takeaways
- For listening to text or articles, built-in voices (Edge Read aloud, Narrator, Mac Speak selection, Android Select to Speak) cost nothing and need no account.
- ElevenLabs Free is 10,000 credits per month and non-commercial only. Commercial use starts at the paid Starter plan.
- Google Cloud lists 4 million free characters a month for Standard and WaveNet voices, and 1 million for Neural2 and Chirp 3: HD. Amazon Polly lists 5 million for Standard voices. Both need a cloud account with billing.
- Kokoro-82M has Apache-licensed weights, which its authors say can be used in production. Piper is GPL-3.0 code, its project says it is intended for personal use and research only, and each voice has its own license, so read the voice card before publishing.
- None of these is a substitute for consent: do not clone or imitate a real person's voice without their permission.
What do you get free, and what is the limit?
The table lists eight options with the limit each provider states itself. "Not stated" means the page we read does not say, not that the answer is yes.
| Option | What you get free | Main limit | Commercial use |
|---|---|---|---|
| Windows: Edge Read aloud and Narrator | Natural sounding voices in Edge Reading mode (F9), Narrator with legacy and natural voices | Reads on screen, no export to an audio file on the pages we read | Not stated on the support pages |
| macOS: Speak selection | Your Mac speaks selected text, shortcut Option-Esc by default | Reads aloud, no download button on the page we read | Not stated on the support page |
| Android: Select to Speak | Reads selected items on screen (Android 11 and up for some steps) | Accessibility feature, not a voiceover maker | Not stated on the help page |
| Kokoro-82M (open source) | An 82 million parameter model you run yourself, pip install kokoro | You need Python and espeak-ng (a separate GPL program), or a Colab notebook | Yes, Apache-2.0 license; the authors say weights can be deployed in production. Credit line: the model card lists CC BY training audio (Koniwa, CC BY 3.0; SIWIS, CC BY 4.0) |
| Piper (open source) | Fast local neural engine, pip install piper-tts, many voices | Code is GPL-3.0; the project looks for maintainers | Not confirmed by the project. VOICES.md: "Piper is intended for personal use and text to speech research only; we do not impose any additional restrictions on voice models." Check each voice's MODEL_CARD |
| ElevenLabs Free | 10,000 credits per month | Non-commercial use only; no commercial license on the Free plan | No. Commercial License starts at Starter ($6 per month) |
| Google Cloud Text-to-Speech | Per month: 4 million characters Standard and WaveNet, 1 million Neural2 and Chirp 3: HD | Developer API with billing enabled; charged past the free amount | Pricing page states no restriction; read the service terms |
| Amazon Polly | Per month: 5 million characters Standard; 1 million Neural, 500 thousand Long-Form, 100 thousand Generative for the first 12 months | Developer API on an AWS account; charged past the free amount | Pricing page states no restriction; read the AWS service terms |
The Polly pricing page also describes a 6-month free plan with up to $200 in credits for new AWS customers who sign up after July 15, 2025, which can replace the free tier above; check which applies to your account. Prices after the free amount, for reference from the same pages: Polly Standard $4.00 per 1 million characters and Neural $16.00; Google Chirp 3: HD $30 per 1 million characters.
Which free text to speech should you use for a video voiceover?
If the video earns money, the license decides, not the voice quality. A free ElevenLabs account is non-commercial by its terms, so use a paid plan, or choose Kokoro, whose weights are Apache licensed, or a cloud allowance under the provider's terms.
Kokoro needs setup but no account and has no character cap beyond your computer's speed. On our own server the Kokoro voices af_heart and am_michael made about 15 seconds of audio in about 4 seconds. For a quick social clip without a setup step, the text to speech built into editing apps is easier; see our guides for CapCut and TikTok.
For Piper, check the license on the specific voice you choose. Its VOICES.md says: "Piper is intended for personal use and text to speech research only; we do not impose any additional restrictions on voice models." It also warns that some voices may have restrictive licenses. The project therefore does not confirm commercial use: read each voice's MODEL_CARD before publishing a monetized video.
Which free option is best for accessibility and reading aloud?
Use the built-in tools. Edge Reading mode has Read aloud with speed control and follow-along highlighting. Windows Narrator offers natural voices, macOS Speak selection speaks any selected text, and Android Select to Speak reads selected items. They work on any page or document without an account or a character cap.
Piper is also used for accessibility: its repository lists NVDA, a screen reader, among the projects that use it.
Which free option is best for quick drafts and scripts?
Pick the one you can open in under a minute. ElevenLabs Free gives the most natural voices for a short test (10,000 credits per month), but the output is non-commercial. For a rough read-through of a script, Edge Read aloud or Mac Speak selection is enough and costs nothing.
A cloud allowance is large (millions of characters), but you must create a cloud account and set up billing first, so it suits developers more than someone who wants one audio file today.
Is free text to speech good enough, and where does it fall short?
Quality depends on the voice type. Google lists Standard, WaveNet, Neural2 and Chirp 3: HD as separate voice families at different prices, and Polly does the same with Standard, Neural, Long-Form and Generative. The largest free allowances on both pages are for the Standard voices, and the newer voice families have smaller free amounts, so the voices with the most free text are not the most advanced ones the providers sell.
Open models are also limited. Kokoro is an 82 million parameter model, built to be light and fast, and the repository describes text to speech, not voice cloning. Free plans also often exclude things a creator needs: ElevenLabs lists Instant Voice Cloning and the Commercial License starting at Starter.
Can you use free text to speech to copy a real person's voice?
Not without that person's consent. Rules on voice likeness vary by country; this is not legal advice. Cloning a real person's voice without permission can break a service's rules and the person's rights. Use the stock voices, a voice you have recorded yourself, or a clone where the speaker has agreed in writing.
Is there an unlimited free text to speech?
Open source on your own machine is the closest: Kokoro and Piper have no per-character quota. The cost is your own computer time and setup. Web services that advertise free speech can still cap length or restrict commercial use, as ElevenLabs does, so read the terms page before you publish.
Questions people ask
Is ElevenLabs free text to speech free for commercial use?
No. The ElevenLabs Terms of Service (updated October 9, 2026) say free users may only use the services for non-commercial purposes. Its pricing page lists the Commercial License starting at the Starter plan.
How much text can I convert free on ElevenLabs?
The Free plan is 10,000 credits per month, as stated on its pricing page. The page does not give a minutes figure for the Free plan, so we do not convert credits to minutes.
Is Kokoro free to use commercially?
Its repository and model card say the weights are Apache licensed and can be deployed anywhere from production to personal projects. The code is Apache-2.0 on GitHub. The model card also has a Creative Commons Attribution section listing training audio from Koniwa (CC BY 3.0) and SIWIS (CC BY 4.0); read it for the credit those licenses ask for. Kokoro needs espeak-ng, which has its own license.
Is Piper free for commercial use?
The project does not confirm it. The Piper code is GPL-3.0, and its VOICES.md says: "Piper is intended for personal use and text to speech research only; we do not impose any additional restrictions on voice models." Voices have their own licenses in a MODEL_CARD file, and some may be restrictive, so check the voice you use before a monetized video.
How many free characters does Google Cloud Text-to-Speech give?
Per month the pricing page lists 4 million characters for Standard and WaveNet voices and 1 million for Neural2 and Chirp 3: HD, then charges per character.
How many free characters does Amazon Polly give?
Its pricing page lists 5 million characters per month for Standard voices, and for the first 12 months 1 million for Neural, 500 thousand for Long-Form and 100 thousand for Generative.
Can the built-in Windows, Mac or Android voices make an audio file?
The support pages we read describe reading text aloud on screen. They do not describe saving an audio file, so use an engine such as Kokoro or Piper if you need a file you can edit.
Related
Sources and how we checked
- ElevenLabs pricing: Free plan, credits, Commercial License (checked 2026-10-20)
- ElevenLabs Terms of Service (non-EEA), last updated 09 October 2026 (checked 2026-10-20)
- Google Cloud Text-to-Speech pricing: free usage limits per month (checked 2026-10-20)
- Amazon Polly pricing: free tier (checked 2026-10-20)
- Kokoro repository (hexgrad/kokoro) and Apache-2.0 license (checked 2026-10-20)
- Kokoro-82M model card on Hugging Face (license: apache-2.0) (checked 2026-10-20)
- Piper repository (OHF-Voice/piper1-gpl), GPL-3.0 (checked 2026-10-20)
- Piper voices and licenses (VOICES.md) (checked 2026-10-20)
- Microsoft Support: Use Reading mode and Read aloud in Microsoft Edge (checked 2026-10-20)
- Microsoft Support: Complete guide to Narrator (checked 2026-10-20)
- Apple Support: Have your Mac speak text that appears on the screen (checked 2026-10-20)
- Android Accessibility Help: Use Select to Speak (checked 2026-10-20)
We fetched each provider's pricing, terms or repository page on October 20, 2026 and copied limits as written. The chart is drawn by code from the table numbers. The Kokoro timing is from our own server run (about 4 seconds for 15 seconds of audio). We did not test the voices against each other, so we make no quality ranking. ElevenLabs, Google, Amazon, Microsoft, Apple and Android are trademarks of their owners; Koldflux is not affiliated with or endorsed by them.