لهجات أصيلة
/ʕ/Dialect-native voices
Multi-dialect speech today, each with native rhythm and vowel color, not a re-accented standard.
Dialect-native Arabic voice - synthesis & recognition
Natiq — an Arabic voice model built around the language as it is spoken.
mithal builds Natiq: an Arabic voice model that turns text into natural speech and transcribes it back - across Masri, Khaleeji, Hijazi and MSA, each with its own vowels, rhythm and sound.
text-to-speechtranscription (ASR)voice cloning from secondsmulti-dialectArabic-first modeling
§ 01 - Listen
Same meaning. Four mother tongues.
Pre-rendered samples — nothing is generated live here.
text to speech
Modern Standard
كان العمل اليوم مزدحمًا جدًا، كان لديّ اجتماع تلو الآخر، ولم أتمكن حتى من أخذ استراحة.
settings
dialect
voice
speaker encoder
cfg scale1.40
dry multiplier0.80
الفصحى
current target
§ 02 - The model
Not a multilingual model with Arabic bolted on. A speech model built around the language's structure.
استنساخ الصوت
/q/Voice cloning
Capture a speaker from a few seconds of reference audio with the speaker encoder.
~6s reference -> matched timbre
التشكيل التلقائي
/h/Diacritization-aware
Reads bare, undiacritized text and infers the vowels the way a native speaker would.
علم -> عِلْم / عَلَم
تبديل لغوي
/sh/Code-switching
Moves between Arabic and English mid-sentence without breaking prosody or pace.
AR + EN, one breath
تحكّم بحثي
/s/Research-grade control
CFG scale, candidates, and sampling controls. Real parameters, not one quality slider.
cfg · candidates · min-p · dry
بنية عربية أولاً
/dh/Arabic-first design
Tokenization, conditioning, and evaluation choices shaped around Arabic rather than inherited from a general multilingual system.
language-specific by design
§ 03 - Platform
A first look at the workspace behind the voice.
Write in Arabic, choose a dialect and voice, then shape the result from one focused interface.
NATIQ V1
أهلًا بيكم، دي أول مرة تسمعوا فيها صوت ناطق باللهجة المصرية.
Try an example
MSA
Egyptian
Emirati
Saudi
Generate Speech
§ 04 - Reading the unwritten
Arabic leaves the vowels out. The same three letters, three different words.
Natiq infers the missing vowels from context, then says the sentence the way a native speaker would.
bare text
علم
عِلْم
ilmknowledge
العِلْمُ نورٌ يهدي صاحبَه إلى الحقيقة.
عَلَم
alama flag
رُفِعَ العَلَمُ عاليًا فوق السارية.
عَلَّمَ
allamahe taught
عَلَّمَ المعلمُ تلاميذَه بصبرٍ وحبّ.
§ 05 - Codeswitching
Arabic and English can share the same utterance without breaking the rhythm of the sentence.
ممكن تعملليforwardللـemail؟
§ 06 - And it listens
The same model runs in reverse - dialect speech in, transcribed text out.
upload audio · Egyptian
Drop an audio file here
WAV, MP3, or M4A
egyptian-voice-note.wavuploaded
transcribed text
الكلام المصري له إيقاع خاص ومميّز جدًا.
§ 07 - Film
See Natiq in action.
A short introduction to dialect-native Arabic speech, presented in Arabic.
About mithal
Mithal is an Arabic language research project built by Ruliad, a UAE-based AI research lab.
Our first model is Natiq: an Arabic speech system developed around the language as it is actually spoken—across dialects, registers, and mixed-language use.