A text to speech API lets your app, website or agent turn written text into spoken audio in real time or as files. APIs differ in latency, voice range, languages, timestamps and where they can run. Choose by whether you need live conversation or recorded audio. Ten options follow, in no particular order.
Picking a voice API is a product decision as much as a technical one. Latency decides whether a voice agent feels natural, voice quality decides whether users trust it, and the API's terms decide what you can build.
The ten APIs below cover the big cloud providers, real-time voice specialists and open models. They're listed alphabetically and in no particular order, and every description comes from the vendor's own website, read this week. If you only need audio files and not code, my notes on converting text to speech with AI for free are simpler.
Text to speech APIs at a glance
I checked the free plan column on each tool's own website, as of Oct 2026. Yes means the site says the tool, or a plan or version of it, is free. Trial only means a time-limited trial or trial credit. Not stated means the pages I read named neither, even where there's a button inviting you to start, so check the vendor's page before you sign up.
| Tool | Type | Good for | Free plan |
|---|---|---|---|
| Amazon Polly | Managed cloud text to speech service | Apps already running on AWS | Yes |
| Cartesia | Real-time TTS API | Low-latency voices for live agents | Not stated |
| Deepgram | Speech API with context-aware TTS | Voice agents that handle interruptions | Not stated |
| ElevenLabs | AI audio API platform | Text to speech plus other audio APIs | Not stated |
| Google Cloud Text-to-Speech | Cloud text to speech API | Voice interfaces inside Google Cloud apps | Trial only |
| Neuphonic | TTS API with open speech models | Running open voice models yourself | Yes |
| OpenAI | Speech endpoint in a model API | Adding speech to apps built on its models | Not stated |
| Rime | Conversational voice models | Support and booking calls | Not stated |
| Smallest.ai | Fast TTS API | Audiobooks and podcasts in many languages | Not stated |
| Unreal Speech | Text to speech API | Long audio with word timestamps | Yes |
Ten text to speech APIs to compare
Amazon Polly
Best for: Teams on AWS who want a fully managed service that converts any text into an audio stream on demand.
Cartesia
Best for: Live voice app builders who want expressive voices that read order numbers exactly, in 44 languages from one model.
Deepgram
Best for: Teams building enterprise voice agents who want low-latency TTS that handles interruptions and carries context across a conversation.
ElevenLabs
Best for: Product teams who want text to speech, speech to text, voice agents and music from one set of audio APIs.
Google Cloud Text-to-Speech
Best for: Google Cloud customers who want natural voices for app voice interfaces, with free credits for new customers to try it.
Neuphonic
Best for: Builders who want open speech models they can inspect, adapt and deploy on their own terms, plus control of pronunciation.
OpenAI
Best for: Developers already building on its models who want speech and realtime voice features from the same platform.
Rime
Best for: Enterprises who want voice models made for human conversation in customer support, healthcare, banking and hospitality calls.
Smallest.ai
Best for: Startups who want a fast TTS API for audiobooks, podcasts and more across over 70 languages, accents and dialects.
Unreal Speech
Best for: Developers who want per-word timestamps and long audio files, streamed over a websocket with a free API key to start.
Match the API to the product
- Apps already running on AWS: Amazon Polly.
- Real-time agents that need fast first audio: Cartesia or Deepgram.
- One vendor for speech, transcription and music: ElevenLabs.
- Voice interfaces on Google Cloud: Google Cloud Text-to-Speech.
- Open models you can run yourself: Neuphonic.
- Support and booking call voices: Rime.
- Long audio with word timestamps: Unreal Speech.
- Audiobooks and podcasts in many languages: Smallest.ai.
The EU wants AI voices to be identifiable
The European Commission's AI Act page says providers of generative AI have to ensure that AI-generated content is identifiable. It also says people using AI systems such as chatbots should be made aware that they're interacting with a machine.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
If your product speaks to people in the EU, decide how you'll tell them the voice is AI before you choose an API. Check whether the vendor offers watermarking or detection, and add a clear spoken or on-screen notice to your own product.
Six things to test before you commit
- Time to first audio on a real network, not a demo page.
- How it reads numbers, dates, addresses and order codes.
- Whether it supports the languages and accents you need.
- Streaming, word timestamps and the audio formats you use.
- Limits on requests and character counts.
- Where your data is stored and whether it's used for training.
Real-time or batch?
Real-time APIs stream audio as it's generated, which voice agents and assistants need. A delay of even a second makes a conversation feel broken.
Batch APIs create files from longer text, which suits audiobooks, articles and course content. Here, quality and pronunciation control matter more than speed.
Signs you picked the wrong API
- Users talk over the voice. Latency is too high.
- Order numbers come out wrong. Test number handling.
- Accents sound off. Try region-specific voices.
- Usage limits bite. Check the quotas before launch.
Do you make a text to speech API? Tool makers can ask to be considered for a featured place on this list.
My plain advice on choosing a voice API
Build a tiny test harness first. Send the same twenty sentences, including numbers, names and a long paragraph, to three APIs and score the results with real users. It takes an afternoon and saves you rebuilding your product around the wrong voice.
If you're new to model APIs in general, my explainer on what a model API is covers keys, requests and usage limits in plain English.
Frequently asked questions
What is a text to speech API?
It's a service your software calls to turn text into spoken audio, either streamed in real time or saved as a file.
Which text to speech API is fastest?
It depends on your network and settings. Test time to first audio yourself with your own text.
Is there a free text to speech API?
Some providers name a free plan, free tier or open models. Check the free plan column above and each vendor's site.
Do I need to tell users the voice is AI?
In the EU, the AI Act brings transparency duties. Telling users plainly is good practice everywhere.
Can I use a cloned voice through an API?
Many APIs support voice cloning. Only clone voices you own or have written permission to use.
Can I run text to speech on my own servers?
Some vendors offer open models or on-premises deployment, which helps when data can't leave your systems.
