Voice Design
Create custom AI voices by describing the characteristics you want, like tone, age, and speaking style.
Voice Design lets you create a voice from a text description instead of an audio sample. Describe the voice you want and the model generates it for you.
How It Works
Open the Design tab in the Lab. It has three panels:
- Voice description (left) - Describe the voice you want to create
- Text to speak (center) - Enter the text the new voice will read and pick a model
- Advanced settings (right) - Fine-tune how the model generates speech
Below the panels is a History section with every voice you have generated. The player bar along the bottom edge plays the latest result.
Describing a Voice
The Voice description panel changes with the model you pick.
Qwen3 and VoxCPM 2 take a free-text description under Describe the voice. Be as specific as you like. Some examples:
- "A warm, gentle young female voice with a soft tone"
- "Deep male voice, 40 years old, authoritative and confident"
- "Energetic and enthusiastic, speaking with excitement"
- "Calm and soothing, like a meditation guide"
The more descriptive you are, the closer the result will match what you have in mind. Try including details like age, gender, emotion, pace, and style. Under the box, Need inspiration? Try an example offers ready-made descriptions you can click to fill in.
OmniVoice uses menus instead of free text: Gender, Age, Pitch, English Accent, and a Whisper switch. Leave any of them on Auto to let the model decide.
DramaBox reads a description with the dialogue in quotes, for example A woman speaks warmly, "Hello, how are you?" She laughs, "It's so good to see you!". The center panel shows What will be spoken, and Try an example loads ready-made scenes. You can also pick a saved voice under Voice reference (optional) to steer the sound of the result.
For detailed prompting techniques, example prompts, common mistakes, and an iterative workflow, read the Voice Design Prompting Guide.
Models
Click the model name next to Model, above the text box, to switch models.
| Model | How you describe the voice |
|---|---|
| OmniVoice | Menus for gender, age, pitch, accent, and whisper |
| Qwen3 | Free-text description |
| VoxCPM 2 | Free-text description |
| DramaBox | Description with quoted dialogue, for expressive performance-style reads |
Saving a Designed Voice
Once you generate a voice you like, click Save Voice at the bottom of the Voice description panel, or Save voice on the item in History. Give it a name and click Save. Click Save & use to save it and jump straight to the TTS tab with the voice loaded.
If the generated clip is longer than 15 seconds, a Trim voice clip window opens first so you can keep the best part.
Designed voices go into the same library as your cloned voices. You'll find them in the TTS tab's Voices panel with a Design badge, and in Audiobooks for long-form generation.
With DramaBox, save the voice from its item in History.
Long Text
When the text to speak is longer than 200 words, the Design tab generates a short sample, saves the voice for you, and offers to Create a project that uses it. The project splits your full text into clean chunks.
Use Cases
Character voices
Design distinct voices for different characters in audiobooks, games, or animations without needing voice actors.
Prototyping
Quickly test how different voice styles work with your script before committing to a specific voice.
Brand voice creation
Design a unique voice that represents your brand for consistent use across all content.