Stability, Similarity and Speed: AI Voice Settings Explained
What the stability, similarity and speed sliders do to an AI voice, when to use the Natural, Narration and Expressive presets, and starting points for common jobs.
Short answerStability controls how consistent or varied an AI voice sounds: higher is steadier, lower is more expressive. Similarity controls how closely the output sticks to the original voice. Speed controls pace. For most scripts, start with a preset (Natural, Narration or Expressive) and change one slider at a time only if something sounds wrong.
Open Voice Settings in Voice Labs and you will find three sliders and three presets. They look technical, but each one answers a simple question about the performance you want.
- Stability: should the voice be steady, or should it have range?
- Similarity: how closely should it stick to the voice’s original character?
- Speed: how fast should it talk?
Stability: steady or expressive
The app describes this slider as controlling voice consistency and variability, and its own tip adds that higher values create more consistent output but may sound less natural.
Think of it as how much freedom you give the performer.
Higher stability keeps tone, pitch and energy level from start to finish. Use it when a listener will be with the voice for a while, or when clarity matters more than character: lessons, explainers, product walkthroughs, audiobook narration.
Lower stability lets the voice move. Lines differ from one another, emphasis is stronger, and emotion comes through more. Use it for character dialogue, storytelling with drama, comedy, and short pieces that need energy.
There is a cost at each extreme. Very high stability can drift towards monotone. Very low stability can be unpredictable, so two generations of the same line may sound quite different. The middle of the slider is the default for a reason.
Similarity: close or loose
The app describes this as how closely to match the original voice.
Every voice in the library has a recognisable character. Similarity decides how firmly the generated speech holds on to it. Higher values keep the result close to the voice you picked. Lower values allow a softer, more generic interpretation.
Most of the time you chose the voice because you liked it, so the default, which sits towards the higher end, is the right place. Lower it only if a voice sounds harsh or exaggerated on a particular script.
Speed: pace and tempo
The app’s label is speech rate and tempo control, and the default is 1.0x.
- Slow down for meditations, bedtime stories, complex explanations and listeners who are learning the language.
- Speed up for short-form video, where pace holds attention, and for ads with a lot to say in a short time.
A warning from the edit suite: speed is the wrong tool for making a script fit. If your 30-second ad runs 36 seconds, cut a sentence. A rushed voice sounds less trustworthy than a calm one saying slightly less.
The three presets
Presets set the sliders for you. These are the app’s own descriptions.
| Preset | The app says | Reach for it when |
|---|---|---|
| Natural | Balanced settings for natural speech | You are not sure. It suits most voiceovers and conversational scripts. |
| Narration | Stable and clear for reading content | The piece is long or instructional: chapters, lessons, documentaries. |
| Expressive | More variation for dynamic content | The piece is short and needs energy or character: hooks, ads, dialogue. |
A sensible routine is to generate your first paragraph with Natural, and switch preset only if you can say what is wrong. “Too flat” points to Expressive. “Too wobbly” points to Narration.
Starting points by job
These are places to begin, not rules. Voices differ, so always listen.
| You are making | Preset | Then try |
|---|---|---|
| A TikTok, Short or Reel | Natural | Expressive, or Speed up slightly |
| An audiobook chapter or story | Narration | Speed down slightly for bedtime stories |
| A lesson or training module | Narration | Speed down for beginners |
| A meditation or sleep script | Narration | Speed down noticeably; choose a soft voice |
| An ad or promo | Expressive | Natural for premium or calm brands |
| Character dialogue | Expressive | Lower Stability further for big personalities |
| A podcast intro | Natural | Expressive if it sounds too formal |
How to tune without getting lost
- Fix the script first. Most delivery problems are punctuation or sentence-length problems. See how to write scripts for AI voices.
- Test on one paragraph, not the whole script.
- Change one thing at a time. If you move two sliders, you will not know which one helped.
- Listen on the device your audience uses. For most people that is a phone speaker or earbuds.
- Write down what worked. Voice name, preset, three slider values. Reuse them for every part of the project so the pieces match.
- Use Reset if you have wandered too far. The Voice Settings screen has a Reset button that returns the sliders to their defaults.
When the settings are not the problem
If you have tried all three presets and the read is still wrong, the answer is usually a different voice. A warm storyteller will never make a convincing drill sergeant, whatever you do to the sliders. Go back to the library, preview a few alternatives against the same line, and start again from Natural.
