Sign inContact usTry Munsit freeENAR

Tashkīl and Arabic tools

Send plain Arabic. Get it back, vowelled.

One POST adds the diacritics a reader, a learner or a voice needs to say the text as written. On the same key: translation that streams, sentiment and keywords on any transcript.

Get your API keyRead the Tashkīl docs
Docs 
curl -X POST "https://api.munsit.com/api/v1/tashkil/diacritize" \
  -H "x-api-key: $MUNSIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "text": "ذهب الطالب الى المدرسة" }'

Marks that settle the reading.

Unvoweled Arabic leaves the reader to infer the case endings and, now and then, which word it is. The response returns the original and the vowelled text together, so you can show both or diff them.

Tashkeel for Modern Standard Arabic.

# one call; three sentences it settles
curl -X POST "https://api.munsit.com/api/v1/tashkil/diacritize" -H "x-api-key: $MUNSIT_API_KEY" -H "Content-Type: application/json" -d '{ "text": "…" }'

The docs' own sentence

ذهب الطالب الى المدرسة

ذَهَبَ الطَّالِبُ إِلَى المَدْرَسَةِ

dhahaba ṭ-ṭālibu ilā l-madrasati

Four case endings the reader would have to supply.

A news lede

أعلنت الوزارة عن خطة جديدة للتعليم

أَعْلَنَتِ الوِزَارَةُ عَنْ خُطَّةٍ جَدِيدَةٍ لِلتَّعْلِيمِ

aʿlanati l-wizāratu ʿan khuṭṭatin jadīdatin li-t-taʿlīm

Formal wording kept; the joins between words voiced.

An ambiguous word

قرأ الطفل الكتب في المكتبة

قَرَأَ الطِّفْلُ الكُتُبَ فِي المَكْتَبَةِ

qaraʾa ṭ-ṭiflu l-kutuba fī l-maktabati

كتب alone could be four words. The marks pick one.

What it's for.

Three places where the vowels are the difference between text that can be read and text that can only be guessed at.

Narration and speech

Fusha scripts diacritized before synthesis, so the voice reads every ending as written.

Reading and learning tools

Learners see the pronunciation on the page instead of inferring it.

Publishing and review

Formal Arabic normalised before it goes out, in one synchronous call.

Then say it.

Vowelled text is the input speech wants. Three calls take a paragraph from plain text to a narrated clip with every word timed.

  1. text

    Diacritize

    Send the plain Fusha; read data.diacritized_text.

    POST /tashkil/diacritize
    AudiobooksRead as the author wrote it.
  2. speech

    Synthesize

    Feed the vowelled text straight to a voice. Streamed, the first chunk plays before the rest exists.

    POST /text-to-speech/{model_id}
    NarrationPrecise pronunciation on formal scripts.
  3. captions

    Time every word

    The same request with timestamps returns the audio and a character-level alignment.

    POST /text-to-speech/{model_id}/with-timestamps
    Learning appsHighlight each word as it's spoken.

The rest of the Arabic text toolkit.

Four more endpoints on the same key. Three run on a transcript or a recording you already have; one on any text you send.

Translation

Any language to any other, streamed over Server-Sent Events so the first words land while the rest are generated. Source language detected.

POST /translation/stream

Sentiment analysis

Sentiment and keywords run on the transcript, after it is written. Tone, emotions and trends, read with Arabic's cultural context.

POST /audio/{transcriptionId}/sentiment-analysis

Keyword extraction

People, organisations, terms, topics and the numbers that matter, grouped, from a meeting's minutes.

POST /minutes-of-meeting/{transcriptionId}/keyword-extraction

Voice isolation

Keep the voice, drop the noise. Up to 200 MB and 15 minutes per file. MP3, WAV, OGG, MP4, MOV, MKV, WebM.

POST /denoise

Vowel the first sentence. One POST.

Plain Fusha in, the same text back with its marks. Try it for free.

Get your API keyRead the docs