---
title: Arabic text to speech API
description: Generate Arabic speech with POST /v1/audio/speech. Choose a voice, stream PCM or WAV, and configure text normalization and pronunciation.
---

Send `POST /v1/audio/speech` with your API key to receive streamed audio. For voice-agent setup, see [Arabic TTS integrations](/docs/integrations#livekit-tts-configuration).

```bash
curl --fail --show-error https://api.sawtakarabi.ai/v1/audio/speech \
  -H "Authorization: Bearer $SAWTAK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "arabic-tts-1",
    "voice": "<VOICE_ID>",
    "input": "مرحباً بك",
    "response_format": "wav"
  }' \
  --dump-header speech.headers \
  --output speech.partial.wav
```

Keep the `X-Request-Id` from `speech.headers`. Before using the file, [verify completion and finalize its WAV header](/docs/streaming#playing-streaming-wav-audiowav).

## Request fields

| Field | Required | Description |
| --- | --- | --- |
| `model` | No | Defaults to `arabic-tts-1`. |
| `input` | Yes | Nonblank text, at most 10,000 characters and 400 characters per word. |
| `voice` | Yes | A ready voice ID from the [catalog](/docs/voices). |
| `response_format` | No | `pcm` (default) for raw audio, or `wav` for a WAV stream. Case-insensitive. |
| `sample_rate` | No | Integer sample rate from 8,000 to 48,000 Hz; defaults to 24,000 Hz. Read `X-Sample-Rate` in the response. |
| `normalize_text` | No | Boolean, default `true`. Converts numbers, dates, currencies, and units into spoken words. Set to `false` to disable this conversion. |
| `enhance_pronunciation` | No | Boolean, default `false`. Adds Arabic diacritics (tashkeel) to guide pronunciation, using the selected voice’s dialect. Preserves existing diacritics. |

## Text preparation

Text normalization is enabled by default. The service automatically chooses how to read numbers, dates, currencies, and units from the text and selected voice. You can separately enable `enhance_pronunciation` to add Arabic diacritics before speech generation.

Choose a voice with the dialect you want. Send plain text or Markdown; SSML is unsupported. Split text longer than 10,000 characters at sentence boundaries.

See [Streaming audio](/docs/streaming) for playback, saved WAV files, and partial-stream charges. Use [Operations and retries](/docs/operations) for status, cancellation, and duplicate protection.

### Adding pauses

Insert `[short pause]` for a brief pause or `[pause]` for a longer pause directly in your text. Use these lowercase tags between spoken words or sentences:

```text
Hi, good morning! [short pause] This is Maria. [pause] How may I help you today?
```

The voice model produces the silence, so the exact duration can vary. Place tags before more speech, rather than at the end of the text; trailing pause tags are removed. Newlines also add pauses when text normalization is enabled. Sentence-ending punctuation alone does not insert a pause tag.

## Advanced options

Omit these unless you need to adjust synthesis:

| Field | Description |
| --- | --- |
| `cfg_value` | Model guidance strength. Defaults to `2`; numeric values are clamped to 0–10. |
| `temperature` | Sampling randomness. Defaults to `1`; numeric values are clamped to 0–2. |

## Use a custom voice

Create a private voice through [Voice cloning](/docs/voice-cloning), then pass the returned ID as `voice`.

For response failures and retries, see [Errors](/docs/errors) and [Request limits](/docs/limits).

Use text and voices you have the rights and permissions to use, and review the audio before publishing it. See the [Terms of Service](/terms) and [Acceptable Use Policy](/acceptable-use).