How to Make an AI Voice Sound Human, With Real Emotion

Emotion tags, voice effects, direction in plain words, pauses and punctuation: the techniques that turn a flat AI read into a performance, with the exact syntax.

Voicefinder Team
· 6 min read
How to Make an AI Voice Sound Human, With Real Emotion

To make an AI voice sound human, direct it the way you would direct an actor: give each sentence one clear emotion, write the script for the ear, and use pauses, breaths and laughs where a person would. With Voyager 1 you do this with tags in square brackets, such as [upbeat], [hushed] or [sigh], or with a note in plain words like [holding back tears, but smiling]. The voice matters less than you think; the text and the direction do most of the work.

This guide shows the techniques that make the biggest difference, with the exact tag syntax and real audio examples.

Why do AI voices sound robotic?

Most "robotic" AI voice-overs are not caused by the voice itself. They come from three things:

  • Flat scripts. Text written to be read silently (long sentences, passive voice, no questions or exclamations) gets read flatly by humans and AI alike.
  • One tone for everything. A person changes attitude every few sentences. A voice-over that stays in one register for two minutes sounds synthetic even when every word is clean.
  • No air. Real speech has breaths, small hesitations, a laugh, a pause before the important line. Without them, the rhythm is too even.

Each of these has a fix you can apply in minutes.

Hear the difference: one sentence, four directions

Same voice, same words: "Something new starts today. And I wanted you to hear it from me." Only the direction changes.

Hannah · General American · hopeful, with a smile
Hannah · General American · nervous, like right before an exam
Hannah · General American · calm and self-assured
Hannah · General American · moved, close to happy tears

None of these were edited after generation. The difference is entirely in how the line was directed.

How to add emotion to an AI voice with tags

Emotion tags go in square brackets at the start of the sentence they affect, in English, in lowercase, followed by a space. They colour the whole sentence, up to the next full stop.

[upbeat] Good morning, everyone! The sun is out and breakfast is on the house today.
[serious] Before we start, one important note about safety.

Voyager 1 has 32 built-in emotions. Some of the most useful:

TagWhat it doesTypical use
[upbeat]Bright and positiveAds, welcomes, social
[enthusiastic]Strong excitementLaunches, big news
[friendly]Close and warmOnboarding, customer service
[confident]Firm and convincedCorporate, explainers
[serious]Sober, important informationSafety, compliance, news
[serene]Slow and calmWellness, narration
[comforting]"Don't worry, we'll sort it out"Support lines, healthcare
[intrigued]Genuine interest, a real questionPodcasts, documentaries
[hushed]Low, confidential voiceSecrets, suspense
[ironic]Irony, sarcasmComedy, characters
[enigmatic]Intrigue, mystery, an intimate voiceTrailers, thrillers
[threatening]A villain's threatGames, fiction

There is also an intensity control, from natural to maximum, for when you want the same emotion pushed further. Tags are not counted as characters, so directing costs nothing.

Laughs, sighs and breaths: voice effects

Real speech is full of small sounds. Voyager 1 has 18 voice effects that sound exactly where you write them, including [laugh], [chuckle], [belly laugh], [sigh], [gasp], [deep breath], [exhale] and [ahem].

[nervous] Okay. Big pitch tomorrow. [deep breath] No distractions.
And then she said yes. [laugh] I still can't believe it.

Use them sparingly. One effect in a paragraph sounds human; one in every sentence sounds like a parody. Here is the same line read three ways, the last one with laughter:

Hazel · General American · sarcastic, like you saw it coming
Hazel · General American · quietly, so no one can hear
Hazel · General American · laughing

Direct it in your own words

Sometimes no single tag describes what you want. Then write a direction note in square brackets, the way you would brief an actor:

[holding back tears, but smiling] Thank you. Really, thank you for everything.
[like a movie villain, savouring his victory] At last, we meet.

Voyager 1 interprets the note and turns it into emotions, effects and pauses, placed where they belong. A note lasts until the next one, and it never changes your words. Concrete notes work better than vague ones: [warm, like telling a friend some good news] beats [good], and [in a hurry, as if time were running out] beats technical settings like pitch or words per minute.

Sophie · General American · holding back tears, but smiling
Rupert · British RP · like a movie villain, savouring his victory

Start free: 10,000 characters a month

Pauses and punctuation: the rhythm of a human read

Punctuation already creates pauses. Commas, full stops and paragraphs set the rhythm, and a blank line is read as a full stop. When you need an exact pause, write it in seconds:

[serious] This winter, look after the ones you love. [pause:0.8]
[friendly] Breeze Insurance. Always close.
  • Place pauses after punctuation. For a natural breath, 0.3 to 1 second.
  • A pause before the tagline is one of the oldest tricks in radio, and it works here too.
  • Ellipses can suggest hesitation ("I… I don't know"), but do not overuse them.
  • Speed goes from 0.85 to 1.25 ([speed:1.1]). Stay at or below 1.15 for clean diction. Speeding up does not add energy; the script does.

Write for the ear, not the eye

Voyager 1 reads like an actor reads a script: punctuation, sentence length and phrasing weigh more than the voice you choose.

  • Short sentences bring rhythm and urgency; long ones, calm.
  • Questions and exclamations change the intonation. Use them where you would really say them.
  • Second person and imperatives ("Come see us") sound livelier than the passive voice.
  • Emphasis comes from word order and punctuation, not from words in capitals. Keep capitals for acronyms.
  • Write numbers, dates and names the way they are said: "September 30, 2026", not "09/30/2026".

Compare "Our store offers discounts during the weekend." with "This weekend, discounts across the whole store! Come see us." The second one sounds human before any tag is added.

One emotion per spot

Two tags back to back, like [unhappy][angry], do not blend reliably; usually only one is heard. Chain emotions sentence by sentence instead:

[unhappy] I can't believe it. [angry] Late again.

For a mixed feeling, use a direction note: [holding back tears, but smiling].

Styles and voice variants

For a whole piece in one register, use one of the 9 styles: [storyteller], [documentary], [newscaster], [radio spot], [tutorial], [meditation], [movie trailer] and others. Each sets the tone and pace for the read.

Many Spain-Spanish voices also have recorded variants of the same voice, such as Energetic, Conversational, Advertising, Epic or Tender. Start from the variant that is closest to the character, then add tags on top to push or nuance it.

A complete example: a 20-second radio spot

[enthusiastic] Orange Week is here at Lola's Fruit!
[upbeat] Tangerines, juices and fresh-baked cakes, all half price. [pause:0.4]
[urgent] Only until Sunday. [pause:0.6]
[friendly] Lola's Fruit. The good stuff, close to home.

One emotion per line, urgency in a short sentence and a pause before the sign-off. In the studio you can add music and one of 6,000+ sound effects underneath, and give each character in a dialogue its own voice. For more on ads, see AI voice for ads; for long-form, AI audiobook narration.

FAQ

How do I make AI text to speech sound more natural?

Write short, spoken sentences, give each sentence one emotion with a tag, add a breath or a pause where a person would take one, and listen before publishing.

Where do emotion tags go?

At the start of the sentence they affect, in square brackets. Voice effects like [laugh] or [sigh] go at the exact point where they should sound.

Do emotion tags work in other languages?

Yes. Tag names stay in English, and they work across the 55+ languages, including Spanish, Basque, Galician and Catalan.

Are tags charged as characters?

No. Tags, pauses and direction notes are not read aloud or charged.

Can a cloned voice take emotion direction?

Yes. A clone made with AI voice cloning takes tags and notes the same way as a library voice.

Can I use tags through the API?

Yes. Write them in the text; direction notes are enabled with natural_direction: true. See the text to speech API.

Try it on your own script with Voyager 1. Start free: 10,000 characters a month

  • ai voice with emotion
  • text to speech
  • emotion tags
  • voice direction
  • voyager 1
Written by
Voicefinder Team
AI text to speech

Keep reading.

See all 5 stories →