Arabic TTS with Tashkeel Support

Telnyx NaturalHD voices now generate accurate Arabic speech directly from unvowelled script, improving pronunciation, rhythm, and fluency without requiring tashkeel marks.

This update delivers high-quality speech for Gulf Arabic out of the box, while still supporting full diacritic input for developers who need precision.

A layered approach to pronunciation

NaturalHD now treats tashkeel as an optional but supported feature. The TTS engine intelligently infers vowels, stress, and pauses when tashkeel is absent, handling common Arabic text as it appears in chat, IVR menus, or databases. For advanced cases, developers can still send explicit tashkeel or use SSML phoneme tags to control pronunciation.

This enables a 3 layered strategy to balance accuracy and scale.

1. Automatic inference: Default mode for unvowelled text, producing fluent speech across dialects.
2. Explicit tashkeel: Ensures perfect accuracy for names, legal, or religious content.
3. Phoneme-level tuning: Enables fine-grained control through SSML tags for rare or ambiguous terms.

How it works

When text is sent to NaturalHD, the system first checks for existing tashkeel. If none are found, the prosody model predicts missing vowels and adjusts stress and timing for natural rhythm.The voices, haqq, qadir, layla, shams, and sakina, have been tuned for Gulf Arabic. They feature balanced intonation suitable for customer support and conversational agents.

Why this matters

Arabic is one of the most complex languages for speech synthesis.

Without diacritics, even advanced TTS models make unpredictable errors.
With NaturalHD’s new tashkeel handling, developers no longer have to compromise between scale and precision.

You can keep tashkeel in curated content, apply automatic diacritization for user-generated input, and fine-tune pronunciation for high-value words within the same voice pipeline.

Validation and known limitations

Our new Arabic voices - haqq, qadir, layla, shams, and sakina tuned for Gulf Arabic, so you should validate the output against your target dialect.

While automatic diacritization is effective, it remains probabilistic and can fail on rare words. We recommend surfacing confidence scores in your application to handle these edge cases.

For a complete evaluation, test the full pipeline by pairing NaturalHD with an Arabic-capable STT engine like Azure or Whisper-large-v3.

Arabic TTS should sound natural. With this update, every vowel, pause, and inflection is under your control. It's live in the NaturalHD on Telnyx today.