AI & ML
Free Text-to-Speech and Speech-to-Text APIs: 8 Free Tiers Compared
Free text-to-speech and transcription APIs to prototype on: ElevenLabs, Polly, Deepgram, AssemblyAI and more — limits, card rules and resets.
If you want a free text-to-speech API, ElevenLabs and Cartesia each refill a credit allowance every month with no card, and Amazon Polly gives you 5 million characters a month of Standard voices once you have an AWS account with a card on it. For speech-to-text, none of the transcription offers we track refreshes — the free route is a one-time credit, and Deepgram's $200, which does not expire, is the largest of them. Every number below comes from the free tiers hub and was checked against each provider's pricing page on 19 September 2026.
Recurring plan or one-time credit — pick by project stage
The eight speech offers split into two kinds, and they suit different moments in a project.
A recurring plan resets on a schedule. You get a fixed allowance, you use it, and next month you get it again. That fits a side project, an internal tool, or a demo that has to keep working after you stop looking at it — anything with low, steady volume. ElevenLabs, Amazon Polly and Cartesia work this way.
A one-time credit is a budget you spend once. It suits evaluation: pick two or three providers, run the same recordings through each, compare the output against what you need, and choose. The common mistake is claiming a credit the day you find it and letting it sit — sign up when your test data is ready. Deepgram, AssemblyAI, Speechmatics, Gladia and Scaleway work this way. Deepgram states that its credit does not expire; for the others, check the terms when you sign up.
Only one of the eight asks for a card: Amazon Polly, because it runs on an AWS account. A card on file means usage past the free allowance is charged to it, so set a budget alert before you start. The other seven need no card.
Text-to-speech
Three providers below give you a recurring speech-synthesis allowance. Deepgram's credit covers text-to-speech too; it is listed under speech-to-text below.
ElevenLabs
The ElevenLabs free plan gives you 10,000 credits a month, refreshed monthly, with no card. Two conditions come with it. The credits are shared across all of ElevenLabs' products, so anything else you try on the account draws from the same pool as your speech. And commercial use is restricted on the free plan — read the terms before anything generated on it goes into a product people pay for.
How much audio 10,000 credits covers depends on the model; the pricing page has the current rates. It suits prototyping: wiring a voice into a feature, trying voices against your own copy, and deciding whether synthetic speech is good enough before you pay for it. See ElevenLabs in the catalog.
Amazon Polly
Amazon Polly offers 5 million characters a month with Standard voices, and that allowance is ongoing rather than a first-year promotion. Neural, long-form and generative voices are covered too, but only for your first 12 months; the pricing page has those amounts. It is the one speech offer here that requires a card.
The allowance is measured in characters of text, so budgeting is arithmetic: count what you send. It suits teams already on AWS, and utility speech — notifications, read-aloud, announcements — where a Standard voice does the job and the volume is high.
Cartesia
Cartesia gives you 20,000 credits a month — about $1 of usage — with no card, resetting monthly. The hub lists it for real-time voice, which is where to look if you are building something conversational rather than rendering audio files. As with ElevenLabs, credits convert to audio at a rate set by the model. It suits a voice-agent prototype that needs to keep running at low volume.
Speech-to-text
Every transcription offer we track is a one-time credit. Treat them as an evaluation budget: spend two or three on the same set of your own recordings and read the transcripts side by side.
Deepgram
Deepgram gives you $200 in credits, no card, and no expiration. It is the largest balance here, and it covers both speech-to-text and text-to-speech, so a voice pipeline — transcribe the user, generate an answer, speak the reply — can run on one account. Because the credit does not expire, you can sign up now and spend it slowly. What $200 buys in minutes depends on the model; see the pricing page. It suits anyone building a voice feature end to end.
AssemblyAI
AssemblyAI gives you $50 in credits as a one-time grant, with no card. Use it as a second contestant: run the sample you gave the first provider and compare the transcripts line by line.
Speechmatics
Speechmatics gives you $100 in credits, one-time, with no card — twice AssemblyAI's grant and half of Deepgram's. Same use: an evaluation budget. Spend it on audio that looks like production — the accents, microphones and background noise your users actually have — rather than on a clean sample file.
Gladia
Gladia gives you €50 in credits, one-time, with no card. The offer notes that it is strong on European languages, so if your audio is in a European language other than English, put it in the comparison.
Scaleway
Scaleway's offer is a generative-API free tier rather than a speech plan: 1 million tokens, one-time, no card, and it includes 60 minutes of audio transcription. It suits a project that already needs a language model and wants a first pass at transcription on the same account. It sits with the other model API free tiers on the hub. For an LLM on a tier that refills or never expires, see free AI API tiers that last.
All eight side by side
| Provider | Type | What's free | Card | Resets |
|---|---|---|---|---|
| ElevenLabs | Text-to-speech | 10,000 credits / month | No | Monthly |
| Amazon Polly | Text-to-speech | 5M characters / month, Standard voices | Yes | Monthly; other voices first 12 months only |
| Cartesia | Text-to-speech | 20,000 credits / month (~$1) | No | Monthly |
| Deepgram | Speech-to-text + TTS | $200 credit | No | One-time, no expiration |
| AssemblyAI | Speech-to-text | $50 credit | No | One-time |
| Speechmatics | Speech-to-text | $100 credit | No | One-time |
| Gladia | Speech-to-text | €50 credit | No | One-time |
| Scaleway | Generative APIs | 1M tokens, incl. 60 min transcription | No | One-time |
Also free: image, video and 3D generation
The same hub tracks generation plans outside speech. These are the ones that refill on a schedule, and none of them asks for a card:
- Bing Image Creator — 15 fast-mode creations a day; slower generations stay unlimited.
- Leonardo.Ai — 150 tokens a day, reset every 24 hours; unused tokens do not roll over.
- Recraft — daily free credits for raster and vector images, with up to 3 uploads a day and 2 images per generation.
- Meshy — 100 3D generation credits a month, reset on the 1st; credits do not roll over.
- fal — 5 video generations every 24 hours on MiniMax H3 Max, a single model.
- Vidu — trial credits plus a daily login bonus; the amounts are not published.
- Suno — for music rather than pictures: 50 credits a day, non-commercial use only on the free plan.
The hub records the free allowance, not the access route, so confirm a plan includes API access before you build on it. Stability AI (25 API credits) and Runway (125 credits) are one-time grants, so they are left off this list; both are in the image, audio and video section with the rest.
FAQ
Is there a free text-to-speech API with no credit card?
Yes. ElevenLabs (10,000 credits a month) and Cartesia (20,000 credits a month) both refresh monthly and need no card. Deepgram's $200 one-time credit covers text-to-speech too, also without a card. Amazon Polly is the exception: its free tier needs an AWS account with a card.
Is there a free speech-to-text API that resets every month?
Not among the offers here. Every transcription offer is a one-time credit: Deepgram $200 with no expiration, Speechmatics $100, AssemblyAI $50, Gladia €50, and Scaleway's generative-API tier with 60 minutes of transcription. For ongoing transcription at low volume, Deepgram's non-expiring balance is the closest thing to a free plan.
How many minutes of audio do the free credits cover?
The offers are stated in credits or dollars, not minutes, and the rate depends on the model you call — so the conversion is on each provider's pricing page. The one exception is Scaleway, whose free tier states 60 minutes of transcription.
Can I use the ElevenLabs free tier for commercial projects?
The free plan restricts commercial use, so treat it as a place to prototype and check the current terms before you ship. The 10,000 monthly credits are also shared across all ElevenLabs products, so they go faster than the headline suggests if you use more than speech.
Do these free tiers expire?
The monthly plans — ElevenLabs, Cartesia and Polly's Standard voices — keep refreshing. Polly's neural, long-form and generative voices are free for your first 12 months only. Deepgram's credit has no expiration. AssemblyAI, Speechmatics, Gladia and Scaleway are one-time grants: once spent, they are gone.
Every offer here, plus the one-time image and video credits, is on the free tiers hub. For the speech and model tools themselves, browse the AI & ML directory.