Text-to-Speech Studio
Clone voices and synthesize natural speech with AI.
Sign in to get started
No newsletters or spam — your email is only used to sign in. After signing in, you can claim a daily bonus of up to $0.60 to try it out.
Cost per audio
Calculating cost…
About the service
Tag Usage Guide
Pause tags
You can use pause tags anywhere in the text. Various formats are supported:
- Seconds:
[pause:1.5],[pause:2s],[pause:3] - Milliseconds:
[pause:500ms],[pause:1200ms],[pause:800ms]
Welcome to our show! [pause:1s] Today we will discuss interesting topics.
[Alice] I am so excited! [pause:500ms] This will be great.
[pause:2] Let's move on to the main part.
Paralinguistic tags
The service supports tags for non-verbal sounds: breathing, laughter, coughing, sighs, gasps, background noise, lip smacks, and emphasis. These tags are processed directly by the model during speech generation.
| Tag | Effect | Example |
|---|---|---|
<breath> | Breath | I'm tired <breath> let's rest |
<quick_breath> | Quick breath | Running <quick_breath> almost there |
<laughter> | Laughter | That's hilarious <laughter>! |
<cough> | Cough | Excuse me <cough> sorry |
<sigh> | Sigh | Fine <sigh> I'll do it |
<gasp> | Fright or surprise | Oh no <gasp> what happened? |
<noise> | Background noise | Walking <noise> through the forest |
<lipsmack> | Lip smack | Tasty <lipsmack> food |
<accent> | Accent or emphasis | Very <accent> important |
Wrapper tags are also supported:
| Tag | Effect | Example |
|---|---|---|
<laughing>text</laughing> | Speak text with laughter | <laughing>so funny</laughing>! |
<strong>text</strong> | Emphasize text | <strong>very important</strong> |
Language and character switching
Use square bracket syntax for language and speaker changes:
[en:Alice] Hello world
[ru:Bob] Привет мир
[zh:] 你好世界
Recommendations
- Use angle brackets such as
<breath>to avoid conflicts with character names. - Insert tags where they fit natural speech, such as pauses before an answer or sighs when tired.
- Use tags in moderation: 1-2 tags per sentence is usually enough.
- The model controls effect strength; parameters cannot change laughter volume or sigh duration.
What's new
- Improved the quality of TTS generation.
- Reduced the number of incorrect and empty results.
- Faster generation
- Fewer artifacts
- Reduced cost
- Improved tag handling and updated documentation