How to Make an AI Voice Whisper, Sigh or Sound Excited
To make an AI voice whisper, sigh or sound excited, put a tag in square brackets right before the words it should change, for example [whispering] Don't tell anyone. Sounds such as [sigh] play where they stand, and any other tag, such as [excited] or [slowly], styles the words that follow it.
Here is the same line read plainly and then with a tag in front of it:
Voice sable
She leaned closer. Don't tell anyone.
Comparison
She leaned closer. [whispering] Don't tell anyone.
This guide shows what each kind of tag does, where to place it, and how to avoid the mistakes that make tags sound flat or get ignored. The examples use Voxmint voices that support tags.
The two kinds of tag
Voxmint reads square brackets as directions to the voice, not as words to speak. There are two kinds.
| Kind | Examples | What happens |
|---|---|---|
| Sound | [sigh], [uhm], [shh] |
The voice makes that sound at that point in the text |
| Style | [whispering], [excited], [slowly] |
The words after the tag are spoken in that manner, until the next style tag |
A style tag is not limited to a fixed list. Any short phrase in brackets works as a direction, such as [leaning in]. Treat these as hints to the voice rather than switches: a clear, common word like "whispering" or "excited" is more reliable than an unusual one.
One example for each tag
Listen to each tag on its own before combining them. Each clip below plays the plain line first, then the tagged one.
Whispering
Voice sable
I wasn't supposed to tell you this. The door was already open.
Comparison
[whispering] I wasn't supposed to tell you this. The door was already open.
Excited
Voice wren
We just hit ten thousand subscribers. Thank you so much for watching.
Comparison
[excited] We just hit ten thousand subscribers. Thank you so much for watching.
Slowly
Voice hollis
Take a breath. Now press the green button once.
Comparison
[slowly] Take a breath. Now press the green button once.
Sigh
Voice harper
Well, that took longer than expected. Let's try again.
Comparison
[sigh] Well, that took longer than expected. Let's try again.
Hesitation
Voice linus
I think the answer is probably yes.
Comparison
[uhm] I think the answer is probably yes.
Where to put a tag
Placement matters more than the choice of tag.
- Put style tags before the words they change. A tag never affects the words that came before it.
- A style lasts until the next style tag. To return to a normal voice, add another tag such as
[calmly], or start a new line of text with its own tag. - Put sound tags at the start of a line or sentence. With ready-made voices, a sound in the middle of a sentence is often skipped, but at the start of a generation it is voiced.
- Keep tags close to the words. Put the tag on the same line as the words it styles, with a space after the bracket.
Adjacent style tags combine, so [whispering][slowly] asks for both at once.
Combining tags in one script
Because each style tag starts a new segment of the same voice, you can build a short scene in one pass:
Voice kestrel
Everyone stayed quiet. Then the lights went out. Nobody moved.
Comparison
[slowly] Everyone stayed quiet. [whispering] Then the lights went out. [excited] Nobody moved!
That works as an exercise in range, but restraint usually sounds better. One tag per paragraph, used where the story changes, is more convincing than a tag on every sentence.
Try it on your own line
Common problems and fixes
- Nothing seems to change. Check that the tag is in square brackets and sits before the words, not after. Try a plainer word such as "softly" or "excited".
- A tag has no effect on a Vietnamese specialist voice. Those voices do not take tags. Switch to an expressive voice to use them.
- The effect is too strong. Shorten the text that follows the tag, or swap to a milder word like "gently".
- Laughter does not sound right. Laughing is the one sound that rarely works on ready-made voices. Write the line so it carries the amusement in the words, or use
[excited]for a lighter feel. - The sentence sounds choppy. Each style tag begins a new segment, so too many tags means too many small joins. Fewer, longer sections sound smoother.
Tags also count towards the length of your text, so keep them short.
Which voices take tags
Tags work with Voxmint's expressive voices, which cover every language in the catalogue. The Vietnamese specialist voices (those whose names start with vieneu-) do not take tags. If you work in another language, the same bracket rules apply; browse the English voices to hear which delivery suits your project, then add tags in the same way.
FAQ
Can I use any word in the brackets?
Yes for style directions, though common words work most reliably. Sounds such as [sigh] and [uhm] have set names.
Will the same tag sound identical on every voice?
No. Each voice has its own character, so a whisper from a low, intimate voice differs from one by a bright voice. Listen with the voice you plan to use.
Open the Studio, pick a voice, type your first tag in square brackets and hear the difference at Voxmint.
Related guides
- Make Text to Speech Read Numbers, Dates and Acronyms RightFix misread numbers, dates, prices and acronyms in text to speech by writing them as they should be spoken. Hear before-and-after examples.
- Batch Text to Speech: Generate Many Voice Clips at OnceHow batch text to speech turns a list of lines into separate audio clips in one consistent voice. Plan, test and run bulk voiceovers step by step.
- Text to Speech With Word Timings for Captions and SubtitlesLearn how word timings from text to speech help you build synced captions, subtitles and read-along highlights without transcribing your audio again.