Arabic Voices Best Practices

This document provides practical guidance for using SESTEK Arabic Text-to-Speech effectively. It explains how Arabic text is interpreted by the TTS system and offers best practices for optimizing pronunciation, pacing, and speech clarity across different Arabic-language scenarios.

By following these guidelines, you can produce more natural-sounding output and better understand expected system behavior when diagnosing pronunciation issues.

1. Purpose and Scope

Objectives

• Enable high-quality Arabic TTS outputs
• Explain how the TTS system processes Arabic text
• Reduce ambiguity when diagnosing pronunciation issues
• Provide clear explanations for expected system behavior

Scope

• Arabic Text-to-Speech (TTS)
• Modern Standard Arabic (MSA) with dialect-sensitive behavior
• General use and troubleshooting

2. Arabic TTS Processing Overview

SESTEK Arabic TTS follows this synthesis pipeline:

Input Text
    ↓
Text Normalization
    ↓
Phoneme Generation
    ↓
Waveform Synthesis

3. Key Arabic-Specific Challenges

Arabic introduces additional complexity due to:

• Ambiguity caused by absence of written diacritics (tashkeel)
• Multiple valid pronunciations for the same spelling
• Dialect vs. Modern Standard Arabic (MSA) differences
• Mixed usage of MSA and dialectal forms
• Rare or region-specific Arabic names
• Numeric and mixed-language expressions
• Complex number, date, and currency expressions
• Proper nouns, brand, province, street and building names without Arabic equivalents
• Undefined abbreviations
• Missing SSML pauses (no clear separation between phrases and/or numbers)

These factors may lead to pronunciation that feels unexpected without prior normalization or guidance.

4. Pronunciation and Language Customization

4.1. Abbreviations

Abbreviation management allows custom pronunciations for abbreviations. Default system behavior applies unless explicitly customized.

Undefined abbreviations are a common source of unexpected pronunciation.

Tips: Define uncommon or rare abbreviations for precise pronunciations, or replace with synonyms if available.

Examples:

4.2. Names and Brand Names

Proper nouns and brand names without Arabic equivalents are a common source of pronunciation variation.

Tip 1: Prefer Arabic-script equivalents for proper nouns and brand names.

Tip 2: Define abbreviations for brand names without Arabic equivalents.

Tip 3: Expand abbreviations letter-by-letter when needed in context.

Tip 4: Expand abbreviations segment-by-segment for dialect-specific pronunciations.

4.3. Normalization

Normalization converts written text into a spoken-friendly representation before synthesis. This helps improve pronunciation of numbers, dates, and other complex text elements.

4.3.1. Numbers

Long numeric sequences can cause numbers to sound too fast or dense.

Tips: Insert SSML pauses and rewrite numbers in words for natural synthesis. Write numbers in the exact representation form of their value for correct synthesis.

4.3.2. Dates

Unsupported date formats will cause incorrect synthesis.

Tips: Write dates in supported formats for correct synthesis.

4.3.3. Times

Unsupported time formats will cause incorrect synthesis.

Tips: Write time in supported formats for correct synthesis.

4.3.4. Currencies

Tips: Write currencies in supported formats for correct synthesis. Define abbreviations for currency codes.

4.3.5. Addresses

Province, street, and building names without Arabic equivalents may be interpreted unexpectedly. Long numeric sequences may sound dense without pauses.

Tips: Define abbreviations for street or location names. Add pauses for clarity.

4.3.6. Symbols

Example: Input "%" → Default pronunciation: "بِالمِئَةْ"

Tips:

• Define abbreviations for symbols: add "%" → "بِالمِيِّةْ"
• Use diacritics for customized pronunciation: "١٠٠%" → "مِيِّةْ بِالمِيِّةْ"

4.3.7. Emails

Symbols, punctuation marks, and Latin characters in email addresses may be interpreted unexpectedly.

Examples:

Tip: Write non-Arabic handles and symbols in their exact desired spoken form, customized with diacritics based on dialect.

5. Diacritics (Tashkeel) and Dialect Sensitivity

5.1. Automatic Diacritics

SESTEK TTS applies automatic tashkeel by default. Diacritics are required for intelligible Arabic speech. Missing or incorrect tashkeel may result in valid but unintended pronunciations.

Example:

Input: "دخل أحمد البنك"

Possible pronunciations:

• "دَخَلَ أَحْمَدُ الْبَنْكَ"
• "دَخَلَ أَحْمَدُ الْبَنْكْ"
• "دَخَلْ أَحْمَدْ الْبَنْكْ"

Tip: Define diacritics explicitly to guide pronunciation as desired.

Desired input: "دَخَلَ أَحْمَدُ الْبَنْكَ"

5.2. Dialect vs. MSA Expressions

Dialectal expectations may differ from MSA. The same concept expressed differently by dialect:

Tip: Use dialect-relevant expressions or MSA consistently.

Example for Jassim Voice (ar-KW): "الحزَّة"

6. Pauses and Flow Control (SSML)

SSML tags allow customizing pronunciation, intonation, and emphasis. SESTEK supports the most commonly used SSML tags including <break>, <say-as>, <voice>, and audio insertion.

For full details, refer to: SSML Tag Support

Tips:

• Use pauses sparingly - avoid excessive breaks in short sentences
• Place pauses at semantic boundaries
• For long numeric sequences, group digits (3–3–4 or 2–2–2) and add short breaks

Examples:

Without pauses:

"رقم طلبك هو 123456789"

With SSML pauses:

<speak>رقم طلبك هو <break time='300ms'/> 123 <break time='250ms'/> 456 <break time='250ms'/> 789</speak>

Without pauses:

"للتواصل اتصل على 0501234567"

With SSML pauses:

<speak>للتواصل اتصل على <break time='200ms'/> 050 <break time='200ms'/> 123 <break time='200ms'/> 4567</speak>

7. Speech Rate, Volume and Voice Selection

7.1. Adjustable Parameters

• Voice selection - different Arabic voices may vary in clarity. See Supported Languages and Voices for the full list.
• Rate - controls the speaking tempo of the voice
• Volume - controls the base loudness level of the voice

7.2. Recommended Values for Arabic Voices

Start with the values below, then adjust in small steps (±0.05) and validate until the voice sounds natural and intelligible.

8. Emotion, Tone and Phoneme Tags

Emotional delivery cannot be directly controlled via markup. Indirect influence is possible through:

• Sentence structure
• Punctuation
• Strategic pause placement

9. Why Output May Sound Unexpected

9.1. Common Ambiguous Arabic Words

Ambiguous words may receive a valid but unintended reading. Manual diacritics may still be overridden in ambiguous contexts.

Tip: Disambiguate with tashkeel or rephrasing.

9.2. Text Preparation and Cleaning

High-quality Arabic TTS output starts with clean and well-prepared input text.

Tips:

• Split long paragraphs into shorter sentences
• Avoid dense numeric blocks in a single sentence
• Ensure proper spacing between Arabic words

9.3. Mixed-Language Text (Arabic + English)

Excessive language-switching within a single sentence may reduce naturalness.

Tips:

• Avoid keeping English words in Latin characters unless defined in Abbreviations
• Use Arabic-script equivalents (Arabic transliteration) to guide pronunciation

10. Troubleshooting Guide

Word is pronounced incorrectly

Possible causes: Missing diacritics, ambiguous spelling, foreign-origin word.

Actions: Add tashkeel, split the word to modify spelling, or use phonetic alternatives.

Numbers sound too fast or unclear

Possible causes: Long numeric sequences, missing or insufficient pauses.

Actions: Insert SSML pauses, rewrite numbers in words.

Output does not sound dialectal

Explanation: Default behavior prioritizes MSA. Dialectal pronunciation is not explicitly selectable - use dialect-specific vocabulary and expressions to guide the output.

11. Summary

Recommended:

• Use clear, unambiguous dialect-specific Arabic text
• Apply pauses where clarity is critical
• Keep parameters consistent across turns
• Define abbreviations for foreign words, symbols, and currency codes
• Use supported date and time formats

Avoid:

• Mixing dialects within a sentence
• Overusing diacritics
• Expecting emotion control tags
• Passing raw numeric-heavy text without preparation