Skip to content
Talk to usGet Started

AI voice generator
built for people

Type a line and hear it spoken like a person, in about 500 languages and dialects.

Get an API keyTry it now ↗

Try it now

Open the playground ↗
InworldTime to first token –– tok/s
Typical providerTime to first token –– tok/s
Same model, same prompt, two providers.

Pick a use case and press Speak, then read the line aloud.

Try saying "My account was disabled at 8:30 this morning."

ListeningTranscribing

Ask your agent to add Inworld.

Paste it into your agent

Pick from the library, clone a real one, or design a new one

TTS docs ↗

Any voice your app needs, in about 500 languages and dialects. Pick one, describe one in a sentence, or clone one from 3 seconds.

500+languages and dialects
3 sto clone a voice
1 sentenceto design a voice
50 msto first audio, at its fastest
Playground ↗Voice design ↗Voice cloning ↗
voice library · inworld-tts-2

Chapter three. The house on the hill had been empty for years, or so everyone thought.

0:08
{
  "text": "Chapter three. The house on the hill had been empty for years, or so everyone thought.",
  "voiceId": "Beatrice",
  "modelId": "inworld-tts-2",
  "language": "en-GB",
  "instruction": "rich and measured like an audiobook narrator with a hint of mystery",
  "deliveryMode": "BALANCED",
  "enhanceGeneration": true,
  "audioConfig": {
    "audioEncoding": "MP3"
  }
}
Each voice reads one line in its native language, with a direction in plain words in the request's instruction field.

Described as“An upbeat male fitness coach in his thirties, loud and punchy, a little out of breath, pushing you on.”

0:04

Described as“A gentle grandmother with a soft Irish lilt, slow and cozy, reading a bedtime story by the fire.”

0:07

Described as“A calm museum guide with a light Italian accent, low and warm, speaking slowly and with wonder.”

0:07
{
  "text": "Last thirty seconds! Come on, push, push, push. You've got this!",
  "modelId": "inworld-tts-2",
  "voiceDesign": {
    "designPrompt": "An upbeat male fitness coach in his thirties, loud and punchy, a little out of breath, pushing you on."
  },
  "enhanceGeneration": true,
  "audioConfig": {
    "audioEncoding": "MP3"
  }
}
Each voice was made from the description beside it, with no audio, using ad-hoc voice design, which is in preview. To keep a design, save it to your library in the playground or with the Voices API.
The clip · 3.0 seconds

“Okay, you will not believe what just happened at practice.”

0:03
The clone · a new line

“This voice came from a three-second clip. Now it can read anything you type.”

0:04
POST /voices/v1/voices:clone
{
  "displayName": "My voice",
  "languageCode": "en-US",
  "voiceSamples": [
    {
      "audioData": "<base64 of the clip>"
    }
  ]
}

Upload or record a clip in the playground, or send it to POST /voices/v1/voices:clone. You get a voice ID back that works in every request, at the same price per character as a library voice.

The clip (3.0 s) was cloned with the instant cloning API, then the clone read a line it had never heard. The clip itself was made with voice design.

We actually made it. After all that, we made it.

Directionquietly almost to herself still in disbelief

0:06
{
  "text": "We actually made it. After all that, we made it.",
  "voiceId": "Sarah",
  "modelId": "inworld-tts-2",
  "instruction": "quietly almost to herself still in disbelief",
  "deliveryMode": "BALANCED",
  "enhanceGeneration": true,
  "audioConfig": {
    "audioEncoding": "MP3"
  }
}
Voice Sarah on Realtime TTS-2, the direction in plain words in the request's instruction field. Inline brackets are only for a delivery that changes partway through a line. Written direction works on Realtime TTS-2.
English with a Spanish proverb · Ashley

My grandmother always said poco a poco se anda lejos.es-MX Little by little, you go far.

0:06
<lang lang="en-US">My grandmother always said</lang> <lang lang="es-MX">poco a poco se anda lejos.</lang> <lang lang="en-US">Little by little, you go far.</lang>
English with a Japanese dish · Jessica

Order the 親子丼.ja-JP Trust me, it is the best thing on the menu.

0:04
<lang lang="en-US">Order the</lang> <lang lang="ja-JP">親子丼.</lang> <lang lang="en-US">Trust me, it is the best thing on the menu.</lang>

Set language for the whole line, or wrap a phrase in <lang lang="es-MX"> to switch for just those words. The voice stays the same.

On Realtime TTS-2 the language field covers about 500 languages and dialects.

Picking a voice

Audio formats

Download MP3, WAV, OGG Opus, FLAC, PCM, LINEAR16, μ-law or A-law, at any sample rate from 8 to 48 kHz.

Streaming

The streaming endpoint returns NDJSON, one JSON object per chunk with base64 audio in each, so you can start playback on the first chunk.

One voice ID for TTS and speech to speech

A voice you save here works in the TTS API, in speech to speech for live voice agents, and in the OpenAI-compatible endpoint.

Timestamps and lip-sync

Ask for word or character timestamps to drive captions, and for visemes to drive lip-sync. Visemes add no latency.

Custom pronunciation

Write a name in IPA inside the text and the voice says it the way you wrote it.

Up to 30,000 custom voices

The Growth plan keeps up to 30,000 cloned and designed voices, Builder keeps 3,000 and the free plan keeps 100.

From $7 per million characters, up to 16× lower than ElevenLabs

Pricing ↗

Start free, with up to 70 minutes of speech. Paid plans go down to $7 per million characters.

70 minfree to start
42¢an hour of speech
75¢an hour, directed in plain words
up to 16×lower price than ElevenLabs
text to speech · list price per million characters
Realtime TTS-2 FlashInworld$15to $7
Realtime TTS-2Inworld$25to $12.50
Sonic 3.6Cartesia$37.38
Eleven v4 TurboElevenLabs$40
Eleven v4ElevenLabs$80

Cloned and designed voices bill at the same rate as library voices, so a cloned voice costs up to 8× less than a Google instant custom voice ($7 on Realtime TTS-2 Flash on the Growth plan against $60 per million characters; 4× on demand).

The solid part of each Inworld bar is its lowest paid-plan rate.

For teams building for people

Case studies →
LatitudeLatitude

Every story beat, read aloud.

97%of the voice in Latitude's Voyage, on Inworld
up to 5×faster than the closest alternative
Read the story ↗
Ten millionlanguage learners on Talkpal
TalkpalTalkpal

Talkpal's AI teachers speak in Inworld voices.

40%lower TTS cost
80+languages taught by its AI teachers
Read the story ↗
LiveKitLiveKit

Inworld voices in LiveKit Agents, with no second API key.

TTS + STTin LiveKit Agents
One flagturns on Expressive mode
Read the story ↗

From the browser to the API

01Try it in the browser

Open the TTS Playground, pick a voice, type or paste your script, and hit Generate. You can design and clone voices there too.

Open the playground ↗02Get an API key

The On-Demand plan is free to start and includes up to 70 minutes of speech. The same key calls LLM Inference, with open models from $0.07 per 1M input tokens.

Get an API key ↗03Send the same voice ID

The voice you picked in the browser is the voiceId in your request. Copy a sample below and add your key to run it.

TTS quickstart ↗
curl -s https://api.inworld.ai/tts/v1/voice \
  -H "Authorization: Basic $INWORLD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "I cannot believe you remembered.",
    "voiceId": "Sarah",
    "modelId": "inworld-tts-2",
    "instruction": "warm and quietly moved",
    "enhanceGeneration": true,
    "audioConfig": {"audioEncoding": "MP3"}
  }' | jq -r .audioContent | base64 --decode > sarah.mp3
Output

Realtime TTS-2, voice Sarah, the direction in plain words in the instruction field. Wrote sarah.mp3.

0:02
DocsText to speech ↗Quickstart ↗TTS API reference ↗Voice design ↗Voice cloning ↗Languages ↗LLM Inference ↗Pricing ↗

AI voices built for people run on Inworld. Who are you building for?

Questions

What is the Inworld AI voice generator?

The Inworld AI voice generator turns typed text into speech with Realtime TTS-2, in any of 275 library voices or in a voice you design from a sentence or clone from 3 seconds of audio. The voice you pick in the browser is the same voice ID you send from your code, so a voice you choose here works unchanged in your app.

Is the Inworld AI voice generator free?

Yes, the Inworld AI voice generator is free to start on the On-Demand plan, which includes up to 70 minutes of speech and 100 custom voices, with voice cloning and voice design. After that, paid plans take Realtime TTS-2 Flash down to $7 and Realtime TTS-2 down to $12.50 per million characters, from $15 and $25 on demand. At the Growth plan rates, an hour of speech costs about $0.42 on Flash and $0.75 on Realtime TTS-2, at about 1,000 characters a minute.

Which languages does the Inworld AI voice generator support?

The Inworld AI voice generator speaks about 500 languages and dialects on Realtime TTS-2. The 275 library voices are native speakers of 15 of those languages, in 23 locales, and any voice can speak another language. For the closest match in a new language, design or clone a voice in that language.

Can I clone an existing voice?

Yes, the Inworld AI voice generator clones a voice from as little as 3 seconds of audio and uses up to 30 seconds, as a wav, mp3 or webm file. You can clone in the playground or through the API once you confirm you have the rights to the voice. Professional voice cloning, in beta, trains on more audio for a closer match.

Can I design a voice without audio input?

Yes, Inworld AI voice design builds a new voice from a written description of its age, accent, pace and tone, with no audio input. A description can run from 7 to 1,024 characters in an API request, or 1,000 in the portal, and each request returns up to three previews. Save the one you like and use its voice ID like any library voice.

How do I control how a line sounds?

In the Inworld AI voice generator, put the direction for Realtime TTS-2 in the request's instruction field, in plain words, such as quietly, almost to herself, still in disbelief. Inline brackets are only for a delivery that changes partway through a line. Sounds such as a laugh or a sigh go in the text where they happen, on both models. To switch language inside one line, wrap a phrase in a language tag, such as <lang lang="es-MX">poco a poco se anda lejos</lang>, and the same voice says that phrase in Spanish.

How fast is the Inworld AI voice generator?

The Inworld AI voice generator runs on Realtime TTS-2 Flash, which returns its first audio in as little as 50 ms, and Realtime TTS-2, as little as 90 ms at P90. Realtime TTS-2 Flash returns first audio up to 3× sooner than ElevenLabs Flash v2.5. The streaming endpoint sends audio as it is made, so playback starts before the line is finished.

Can I use the voices commercially?

Yes, speech from the Inworld AI voice generator can be used commercially, because every plan includes a commercial license, starting with the free-to-start On-Demand plan, and our service terms leave you the rights to your output. For a cloned voice, you confirm you have the rights to that voice before you clone it. A voice you save in the voice generator keeps the same voice ID in your app through the TTS API, and in speech to speech as the voice of a live voice agent.