Text to Speech turns any script into natural-sounding audio in seconds. You pick a voice, choose a model, add emotion if you want it, and download the result. This guide walks through every control on the page.
What is Text-to-Speech (TTS)?
Text-to-Speech (TTS) converts written text into spoken audio using AI voices. For example, you type "Hello, how are you?" and CAMB.AI reads it aloud in the voice you choose.
Behind the scenes, the model reads your text, works out pronunciation, pauses and intonation from the words and punctuation, and then generates the audio in the selected voice. Clear punctuation helps it sound natural.
Before you start
You need a CAMB.AI Studio account. Every generation uses credits from your plan; you can see your remaining credits at the bottom of the left sidebar.
Want to use your own voice or a custom character voice? Create it first: How to Create a Custom Voice (and Get the Best Results).
Generate speech, step by step
Step 1. Open Text to Speech. In the left sidebar, go to Speech Tools → Text to Speech.
Step 2. Add your text. Type or paste your script into the box that says "Write something and we'll turn it into speech…". Not sure where to start? Click one of the examples under or try using an example (Sports Highlight, Social Media, Daily News, Movie Scene).
Note: picking an example also changes the voice and model to match it. Check the Settings panel before you generate.
Step 3. Choose a voice. In the Settings tab on the right, open Narrator voice. You can pick one of the Recommended voices, click Explore more voices to browse the full Voice Library (including voices in My Voices and Workspace Voices), or click Use your voice to clone a new one.
Step 4. Choose a model. Click the model card and then Show more to see all models.
Model | Best for |
MARS8.1-Flash-Beta (default) | Fast, real-time generation (~250 ms latency) for agents, call centers and live assistants. A good everyday choice. |
MARS8-Instruct | The most expressive model, with director-level emotion control through tags like [whispering] or [laughing]. Built for TV, film, dubbing and post-production. |
MARS-X Instruct (Beta) | The newest instruction-guided model, for maximum fidelity and expressive delivery. In beta. |
MARS8-Pro | Balanced speed and quality (48 kHz). The best all-round choice when speed isn't critical, e.g. translation, dubbing and audiobooks. |
MARS8-Flash | Efficient, low-latency generation. |
MARS8.1-Beta | Best quality and widest language coverage. |
Quick picks by use case:
Voice agents, call centers, IVR and live conversational AI: MARS8.1-Flash-Beta or MARS8-Flash.
Film, TV, expressive dubbing and dramatic scenes: MARS8-Instruct (or MARS-X Instruct to try the newest model).
Audiobooks, long-form narration and translation: MARS8-Instruct.
Ads, social videos and explainers: start with the default, and switch to MARS8-Instruct if you need more emotion.
Step 5. Set the language. Language is set to Auto detect by default, which works for most scripts. Results are best when the voice matches the language of your text. If you pick a language that doesn't match the voice, a red warning appears; choose a matching voice or language before generating.
Step 6. Add emotion (optional). Under the text box, Emotion Controls shows quick tags you can click to insert at your cursor, such as [laughter] or [sigh]. Click Browse all to see every tag available for your model. Tags turn orange when they're recognised.
The default Flash model offers a short set of expressive tags ([laughter], [sigh], [confirmation], [question], [surprise], [dissatisfaction]). For many more emotions, sound effects and speaking styles, plus an intensity scale from Subtle to Extreme, click Switch to MARS Instruct in the panel (or choose MARS8-Instruct in Step 4). More tips: How to use emotion tags.
Step 7. Fine-tune with Advanced Options (optional). Expand Advanced Options in the Settings panel:
Speaking rate: slower or faster delivery (1.00 is normal).
Inference steps: lower for faster results, higher for more fidelity.
Enhance reference audio quality: sharpens the reference voice and reduces distortion. Turn it on when the voice was cloned from a recording that isn't studio quality.
Output format: .mp3 (default), .wav, .flac or .adts.
Step 8. Generate. Click Begin Generating Speech. When it's ready, a player appears at the bottom. Press play to listen, click Regenerate for a new take, or click the download icon to save the file.
Find your past generations
Open the History tab in the right panel to replay or download anything you've generated. Use the search box to find a clip, and the ⋮ menu on any item to Move to folder, Edit Details, Move to workspace or Delete it.
Character limits
The counter under the text box shows how many characters you've used. The limit depends on the plan and model (for example, 3,000 characters with MARS8.1-Flash-Beta and 1,000 with MARS8-Instruct). For longer scripts, split the text into sections, or use Audiobook or Video Narration under Projects.
Check out the pricing page for your plan level limits.
Handy shortcuts
<> Get Code: shows the API code for your current settings.
Clear text (eraser icon) and Reset settings (circular arrow) next to the character counter.
Try as API (top right): generate the same audio from your own app with the CAMB.AI API.
Troubleshooting
The voice sounds wrong or has an accent: check the language matches the voice, and look for the red mismatch warning.
An emotion tag is read out loud: the tag isn't supported by this model. Pick it from Browse all so it turns orange, or switch to MARS8-Instruct.
A word or name is mispronounced: try spelling it the way it sounds, then click Regenerate.
You can't generate or see a credits message: check Remaining Credits in the sidebar and click Buy more credits if needed.
Still stuck? Message us from the chat bubble and we'll help.