Text to speech API: what it does, what it costs and how to choose one

You are building something that has to speak: an app that reads out order updates, a bot, a script that makes a hundred voiceovers overnight. So you look for a text to speech API, or TTS API, and land on price pages quoted in characters, tokens and credits that cannot be compared at a glance. This guide explains what a TTS API does, puts the published prices of six providers on one scale, and works out what an hour of audio costs on each. It also says plainly where TTS Maker stands: we have no self-service API, and we sell API access directly, on request.

What a text to speech API is

A text to speech API is a web address that your program sends text to and gets audio back from. There is no page to open and no button to press. Your code sends the text, the name of a voice and the audio format, together with a secret key that identifies your account. The service returns the speech as a file or as a stream, and bills you for what you sent.

You need one in three situations:

• Your app or website speaks text that keeps changing: order updates, news, lessons, replies from a chatbot.
• You are automating: many audio files have to be made with nobody clicking a button.
• The voice has to answer live: a phone menu or a voice agent, where the delay before the first sound matters and you need streaming.

You do not need one to make voiceovers for your own videos. That case is covered further down.

Five things to check before you pick a TTS API

• Your language and its voices. Do not assume. Amazon Polly's voice list has two Hindi voices, Aditi and Kajal, and no other Indian language apart from Indian English. Google's Chirp 3: HD list has Hindi, Indian English and nine more Indian languages. Sarvam AI, an Indian provider, states 11 Indian languages, Hindi and Indian English among them.
• The unit of the price. Some charge per million characters, some per thousand, some in tokens. OpenAI prices its gpt-4o-mini-tts model at $0.60 per million text input tokens and $12.00 per million audio output tokens, which cannot be set beside a per-character price without a test of your own. Convert everything to one unit before you compare. The table below does that for the per-character services.
• What counts as a character. Google's pricing page says spaces and newlines are counted, and so are SSML tags except the mark tag. Amazon's limits page says SSML tags are not counted as billed characters. On a script full of markup, that difference shows up on the bill.
• The size of one request. By the same Amazon page, Polly takes up to 3,000 billed characters in one ordinary request and up to 100,000 in a background task. OpenAI's gpt-4o-mini-tts page gives a maximum of 2,000 input tokens. A long script has to be cut into pieces and joined again by your code.
• The licence. Read what the provider allows for commercial use and for the voices themselves, and never use any voice to imitate a real person.

Published TTS API prices on one scale

List prices as each provider published them on 7 October 2026, converted to the price of one million characters and of one hour of audio. An hour is taken as 54,000 characters, at the rough figure of 900 characters a minute that this site uses everywhere.

• These are pay-as-you-go list prices, before free tiers, taxes, currency conversion and plan discounts. On the day we read it, ElevenLabs' page also showed temporary discounts on its newest models, which are left out here.
• Prices change. Each provider name in the table links to the page the figure came from. Open it before you budget.
• A lower price does not mean a voice you will like. We cannot hear the voices for you. Run the same paragraph of your own script through each service you are considering.
• TTS Maker is not affiliated with any of these companies.

Is there a free text to speech API?

There are free allowances, and each has an edge. Google's pricing page lists 4 million free characters a month on Standard voices and 1 million on Neural2 and on Chirp 3: HD. Amazon Polly's lists 5 million Standard characters a month, and 1 million Neural characters a month for the first 12 months; read the page for how long each allowance lasts on your account. Microsoft's Azure Speech pricing page lists 0.5 million free characters a month; for its paid rates, check that page, because we could not read them from it on the day of writing. Sarvam AI says new accounts receive free credits.

The free limits on the TTS Maker website and bot are not a free API. They are for people using the tool by hand, and our terms ask you not to send automated or bulk requests to the free tier.

Where TTS Maker stands on an API

TTS Maker (ttsmaker.in) does not have a self-service API. You cannot sign up and generate a key yourself, and there is no public documentation page. We would sooner say so than let you plan a project around a sign-up page that does not exist. What we do have is described on the API page: API access that we set up directly with each customer.

It is a paid service. It is not part of the free limits, and there is no free API tier. There is no price list, because the price depends on the voices and the volume you need: we quote it when you ask.

1. Contact us on Telegram. Use the button on the API page, or open the bot and send /api. Your next message then goes to our team. Do not type the enquiry as an ordinary message without that step: the bot turns any other text into audio.
2. Or write to us. The email address and phone number are on the contact page.
3. Tell us four things: what you are building, which languages and voices you need, roughly how many characters a month, and whether you need finished files or live streaming.

You may not need an API at all

If you are a creator who wants a lot of audio, not a developer who wants an endpoint, three things already work without a line of code, each with a free account:

• The website: 20,000 characters in one conversion, about 22 minutes of speech, and 200,000 characters a day on the free Microsoft voices.
• The audiobook maker: on the audiobook maker page, one EPUB, PDF, DOCX or TXT file of up to 20 MB and 1,000,000 characters becomes MP3 files. That is about 18 hours of audio from a single upload.
• The Telegram bot: send a message of up to 4,096 characters, or a file, and the MP3 comes back in the chat.

And before any integration work, listen. The box below takes the kind of line an app sends to a TTS API.

When an API is the wrong tool

• A few videos a week. Pasting a script into a converter is quicker than writing and maintaining code.
• Words that must not be wrong. Medicine doses, legal notices, prices in an advertisement. Automation removes the person who would have listened first, so keep a human check for these.
• A voice that has to be yours alone. The catalogue voices of any service can be chosen by anyone else.