Your First Voice: The Playground and Voice Library

The TTS Playground

The quickest way to hear ElevenLabs in action is the Text to Speech playground. It is a single-screen tool: a text box on the left, a voice selector at the top, and a Play button. Type a sentence, pick a voice, click play. Your first generation takes about five seconds.

The playground is where you will do most of your exploratory work — testing voices against your script before committing to a full project. It is also where you access the core voice settings that control how a voice performs.

Voice Settings: Stability, Similarity, and Style

Every voice has three sliders that shape its output:

• Stability — Controls how consistent the voice sounds across sentences. High stability = more uniform and predictable. Low stability = more varied and expressive, but sometimes erratic. For narration, keep this at 60–75%. For dialogue or character voices, try 40–55%.
• Similarity — How closely the output resembles the original voice sample. Set this above 75% for cloned voices. Lower values let the model take more creative liberties.
• Style Exaggeration — Amplifies the stylistic qualities of the voice. Use sparingly — values above 50% can introduce artefacts. A value of 0 is a safe default.

These settings interact with each other and with the model you select. The same slider values will produce different results on Eleven v4 versus Eleven Flash v2.5. Experiment on short clips before running a long generation.

The Voice Library

ElevenLabs hosts a Voice Library of over 3,000 community-contributed voices. These are voices that other users have generated or shared, spanning dozens of languages, accents, ages, and styles. You can browse and use any shared voice in the library without creating your own clone.

To find a voice that fits your project, filter by:

• Language — Filter to the target language of your content.
• Gender and age — Narrow to the demographic that fits your project.
• Use case — Tags like "narration," "social media," "news," "characters," or "audiobooks" help you find voices trained for a specific context.
• Accent — Filter by regional accent within a language.

Each voice in the library has a play button so you can preview it without selecting it. Click the bookmark icon to save a voice to your collection for quick access later.

Premade vs. Cloned vs. Designed Voices

Try This

Open the ElevenLabs playground and type a 50-word paragraph about a topic you know well. Generate it with three different voices from the library — pick voices with very different descriptions. Then adjust the Stability slider: generate the same text at 30%, 60%, and 80% stability on the same voice. Listen to the difference. You will immediately understand what stability does better than any definition can explain.