# Arabic text to speech with emotion: make an AI voice laugh, sigh and change mood mid-line

> Most Arabic AI voices have one mood: polite. nutq-speak-3 can be excited, amused, tender or angry — and switch between them inside a sentence — because you write the emotion into the script.

- Source: https://nutq.dev/blog/arabic-text-to-speech-with-emotion
- Published: 2026-10-17
- Updated: 2026-10-17
- Author: nutq team (https://nutq.dev)

## Short answer

To make Arabic text to speech sound emotional, write cues in square brackets inside the text: [excited], [amused], [angry], [sad], [scared], [tender] or [surprised] change the delivery from that point on, [laughter] adds a real laugh, [sigh] a sigh, and [pause:800] a timed gap. nutq-speak-3, the default nutq model, performs all of them in 16 Gulf voices, at $0.05 per 1,000 characters.

A flat voice can read a script. It cannot sell a weekend offer, comfort a customer, or make a child laugh at a bedtime story. Until recently that was the ceiling for **Arabic text to speech with emotion**: you could get correct Arabic, but not *felt* Arabic. nutq-speak-3, nutq's newest model, changes that by letting you write the emotion into the script itself.

## What is nutq-speak-3?

nutq-speak-3 is the default nutq speech model for Gulf voices. It does everything the previous expressive model did, with three additions that matter for real productions:

- **A wider emotional range** — the difference between "slightly happy" and genuinely excited.
- **Real laughter** where you write `[laughter]`, not a word that sounds like one.
- **Steadier long reads**, so ads, audiobooks and drama hold the same voice from the first line to the last.

It also reads English words inside Arabic as English, which matters for product names and Gulf code-switching.

## Which emotions can an Arabic AI voice perform?

| Cue | Sounds like | Use it for |
|---|---|---|
| `[excited]` | Energetic, bright | Announcements, offers, good news |
| `[amused]` | Light, smiling, playful | Banter, a joke landing, teasing |
| `[tender]` | Soft and warm | Comfort, bedtime stories, apologies that mean it |
| `[sad]` | Low and heavy | Loss, regret, bad news |
| `[angry]` | Sharp, forceful | Drama, a complaint scene |
| `[scared]` | Tense, shaky | Suspense, warnings |
| `[surprised]` | A sudden lift | Reveals, "really?!" |
| `[neutral]` | Plain read | Switching back |

Two sounds sit beside them: `[laughter]` for a real laugh and `[sigh]` for an audible sigh. Pauses are `[pause]` (half a second), `[pause:800]` in milliseconds, or `[pause:1.5s]`.

## How do you write an emotional Arabic script?

An emotion applies **from where you write it until the next emotion or the next speaker**. So a single voice note can turn:

```text
[amused] هههه لا والله ما توقعت. [pause:700] [sad] بس صراحة… [sigh] ضايقني الموضوع شوي.
```

A radio spot:

```text
[excited] عرض نهاية الأسبوع وصل! [pause:400] خصم خمسين بالمية على كل شي.
[tender] لأنك تستاهل. [pause] [neutral] العرض لين يوم الأحد بس.
```

And a scene with two voices, each line opened with the voice's name:

```text
[Hamdan] [scared] سمعت هالصوت؟ [pause:900]
[Alia] [tender] لا تخاف، هذي الريح بس.
[Hamdan] [surprised] الريح ما تدق الباب!
```

Four rules keep it natural:

1. **One emotion per beat.** Stacking three cues on one sentence sounds acted.
2. **Cues in English, always.** `[حزين]` would be read aloud as a word.
3. **No stage directions** in parentheses or asterisks — they are spoken too.
4. **Write the dialect the listener speaks.** About 13 characters is one second of speech.

## How do you fix a line that misses?

Each take is slightly different, which is a feature: if one line lands wrong, regenerate just that line rather than the whole script. If the emotion is too strong, move the cue later in the sentence or drop it and let the words carry the feeling. If a word is read the wrong way, add diacritics to that one word only. And before a long script, run the free script check in the API or MCP tools: it lists every part, speaker and emotion it found, and flags cues it would skip.

## How do you call it from code?

```bash
curl -X POST https://nutq.dev/v1/speech \
  -H "Authorization: Bearer $NUTQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text":"[excited] وصلت الشحنة أخيرًا! [laughter]","voice":"alia"}' \
  --output line.wav
```

nutq-speak-3 is the default, so you do not need to name it. Pass `"emotion": "tender"` to set the starting mood for a whole text, and let inline cues override it. For live calls, `stream: true` defaults to nutq-speak-2-flash, with first audio about a second after the request; name nutq-speak-3 to stream its wider range at about four seconds to first audio.

## Which voices support emotions?

All 16 Gulf voices — Alia, Hind, Maitha, Amna, Noura, Aisha, Hamdan, Sultan, Faisal and the rest — take emotions with nutq-speak-3. The Egyptian voice reads without them. You can also clone a voice from six to eight seconds of audio and direct it the same way. Hear them on the [voice library](/voices), or type a line on the [Arabic text to speech page](/arabic-text-to-speech).

## What does it cost?

$0.05 per 1,000 characters, the same on every model; emotion costs nothing extra, and silence from pauses is not billed. The $5 of credit at [sign-up](/sign-up) covers 100,000 characters — enough for dozens of ad scripts. Coming from another provider's tags? See the [ElevenLabs alternative for Arabic](/blog/elevenlabs-alternative-arabic-tts), and for production ideas, [Arabic AI voiceover for ads, reels and audiobooks](/blog/arabic-ai-voiceover-ads-audiobooks).

## FAQ

### Can Arabic text to speech express emotions?

Yes. nutq-speak-3 performs seven emotions — amused, excited, angry, sad, scared, tender and surprised — plus real laughter and sighs, in its Gulf Arabic voices. You write them as cues in the text, or set one emotion for the whole request.

### How do I change emotion in the middle of a sentence?

Write the cue where the change starts, for example [amused] ... [sad] .... An emotion applies from where it is written until the next emotion cue or the next speaker, across pauses. [neutral] switches back to a plain read.

### Do I write the emotion cues in Arabic or English?

Always in English, inside square brackets. Arabic inside brackets, such as [حزين], is read aloud as a word.

### Which voices support emotions?

The Gulf Arabic voices with nutq-speak-3 and nutq-speak-2. The Egyptian voice and nutq-speak-1 read without emotion, and laughter is nutq-speak-3 only.

### Does emotion cost extra?

No. Speech is $0.05 per 1,000 characters on every model. The cues themselves are not billed as spoken characters, and pauses are not billed.