# nutq

> Arabic speech API: speech to text that keeps Gulf dialect, and Arabic text to speech with voice cloning.

nutq (نُطق) is an Arabic speech API, built in the UAE, at https://nutq.dev.

- Speech to text (model `nutq-listen-1`): tuned for Gulf and Levantine Arabic, and keeps dialect as spoken instead of rewriting it into Modern Standard Arabic. Also transcribes 13 other languages.
- On narrow-band phone audio it made 40% fewer character errors than the leading Arabic vendor on a genuine call-in. 986 clips were scored against human transcripts (broadcast, spontaneous interview and telephony). It runs 22× faster than realtime: an hour of audio in under three minutes.
- Text to speech (model `nutq-speak-2`, the default): expressive Gulf Arabic with emotions — amused, excited, angry, sad, scared, tender, surprised — set per request or switched mid-text with inline cues such as [sad] and [sigh]. Reads English words inside Arabic as English. 16 ready-made Gulf voices, or clone a voice from six to eight seconds of audio.
- Streaming speech: `stream=true` returns raw 16-bit PCM a sentence at a time, at 8, 16, 22.05 or 24 kHz; on a 25-second paragraph the first audio arrives in 3.7 s instead of 10.6 s for the finished file. The previous model, `nutq-speak-1`, can still be pinned with `model=nutq-speak-1`.
- Other languages (model `nutq-voice-1`): trained voices in English, French, Spanish, Italian, Portuguese, Hindi, Japanese and Chinese.
- Voice agents: a Vapi custom voice at https://nutq.dev/v1/vapi/voice and a Deepgram-compatible live transcription WebSocket at wss://worker.nutq.dev/v1/listen. Guide: https://nutq.dev/blog/arabic-voice-agent-vapi
- Pricing: $0.15 per hour of audio transcribed, $0.05 per 1,000 characters of speech. $5 of free credit on sign-up, no card. Monthly plans: Free (10k credits), Starter $5 (30k credits), Creator $19 (600k), Pro $49 (1.8M), Scale $149 (6M), Business $499 (25M), and Enterprise on request, including on-premise; one credit is one character of speech, and an hour of audio is 3,000 credits. Paying yearly costs ten months for twelve. Failed requests are never billed. https://nutq.dev/#pricing
- On-premise: the same models deployed inside a customer's own network on one 24 GB GPU, with no audio leaving the network and no per-minute fee.
- API: POST https://nutq.dev/v1/transcribe and POST https://nutq.dev/v1/speech, bearer-key auth, JSON or multipart form data. Reference: https://nutq.dev/docs
- AI agents (MCP): Claude, ChatGPT, Cursor, Claude Code, Antigravity and other MCP clients connect to https://nutq.dev/mcp and sign in with nutq (OAuth). Tools transcribe audio or video (text, SRT, VTT) and speak scripts with emotions, sighs, pauses and several voices. Agent Skills: https://nutq.dev/skills/nutq-skills.zip. Setup: https://nutq.dev/app/mcp

## Tools

- [Arabic transcription](https://nutq.dev/arabic-transcription): convert Arabic audio or video to text in the browser, Gulf and Levantine dialect kept. [العربية](https://nutq.dev/ar/arabic-transcription)
- [Arabic text to speech](https://nutq.dev/arabic-text-to-speech): type Arabic and hear it in a Gulf voice, or clone your own. [العربية](https://nutq.dev/ar/arabic-text-to-speech)

## Docs

- [API reference](https://nutq.dev/docs): endpoints, parameters, errors and models.
- [Connect an AI agent (MCP)](https://nutq.dev/docs#mcp): use nutq from Claude, ChatGPT, Cursor and other AI apps.
- [Full blog text](https://nutq.dev/llms-full.txt): every article below, in full.

## Blog (English)

- [How to transcribe Arabic voice notes (WhatsApp, Telegram and phone recordings)](https://nutq.dev/blog/transcribe-arabic-voice-notes.md): In the Gulf, a lot of business happens in voice notes — orders, complaints, instructions. Here is how to turn them into text you can search, forward and act on, in the dialect they were spoken in.
- [On-premise Arabic speech recognition: keeping every recording inside your network](https://nutq.dev/blog/on-premise-arabic-speech-recognition.md): For a bank or a ministry, the first question about Arabic speech recognition is not accuracy. It is where the recordings go. Here is what running the models inside your own network involves.
- [Real-time Arabic transcription over WebSocket (a drop-in for Deepgram clients)](https://nutq.dev/blog/real-time-arabic-transcription.md): If you have already written a Deepgram client, you have already written most of a real-time Arabic transcriber. Change the URL, and here is exactly what comes back — and what does not.
- [How to build a Gulf Arabic voice agent with Vapi](https://nutq.dev/blog/arabic-voice-agent-vapi.md): Most voice-agent stacks fall over the moment a caller speaks Gulf Arabic. Two blocks of Vapi configuration fix both ends of the call — hearing the caller and answering them.
- [Arabic text to speech in n8n: turn any workflow into a Gulf Arabic voice](https://nutq.dev/blog/arabic-text-to-speech-n8n.md): No community node to install and no code step. One HTTP Request node, one credential, and every n8n workflow can speak Arabic — in a Gulf voice, not a newsreader's.
- [Arabic speech to text API: a developer's guide to transcribing Gulf Arabic](https://nutq.dev/blog/arabic-speech-to-text-api.md): One POST request, a file, and Arabic text back — with the dialect intact. What to send, what comes back, what it costs, and where Arabic speech recognition usually goes wrong.
- [Gulf Arabic vs MSA: why most speech models rewrite what your customers said](https://nutq.dev/blog/gulf-arabic-vs-msa-transcription.md): A caller says شحالك and the transcript says كيف حالك. It looks like a correction. It is a change to the record — and for search, analytics and compliance it matters more than it seems.
- [How to transcribe Arabic phone calls accurately (and why most models fail on them)](https://nutq.dev/blog/transcribe-arabic-phone-calls.md): A phone line throws away most of the sound a speech model listens to. Here is what that does to Arabic transcription, what we measured, and how to set up call transcription that holds up.
- [Arabic text to speech with voice cloning: a Gulf Arabic TTS guide](https://nutq.dev/blog/arabic-text-to-speech-voice-cloning.md): Six to eight seconds of someone speaking, the words they said, and the text you want read — that is all a Gulf Arabic voice clone needs. How to get a natural result, and the one trick that fixes most mispronunciations.
- [How to add Arabic subtitles to a video automatically (with dialect intact)](https://nutq.dev/blog/auto-arabic-subtitles-for-video.md): Upload a video, press one button, and get Arabic captions timed to the speech — in the dialect people actually used. Then style them, translate them, burn them in, or export an SRT.

## المدونة (العربية)

- [كيف تفرّغ الرسائل الصوتية العربية إلى نص (واتساب وتيليغرام وتسجيلات المكالمات)](https://nutq.dev/ar/blog/transcribe-arabic-voice-notes.md): في الخليج، يجري كثير من العمل عبر الرسائل الصوتية — طلبات وشكاوى وتعليمات. هكذا تحوّلها إلى نص تبحث فيه وتعيد إرساله وتتصرف بناءً عليه، باللهجة التي قيلت بها.
- [التعرّف على الكلام العربي داخل شبكتك: كل تسجيل يبقى عندك](https://nutq.dev/ar/blog/on-premise-arabic-speech-recognition.md): عند البنك أو الوزارة، السؤال الأول عن التعرّف على الكلام العربي ليس الدقة، بل وجهة التسجيلات. هذا ما يتطلبه تشغيل النماذج داخل شبكتك.
- [التفريغ الصوتي العربي المباشر عبر WebSocket (بديل جاهز لعملاء Deepgram)](https://nutq.dev/ar/blog/real-time-arabic-transcription.md): إن كنت قد كتبت عميلًا لـDeepgram، فقد كتبت معظم مفرّغ عربي مباشر. غيّر الرابط، وهذا بالضبط ما يعود إليك — وما لا يعود.
- [كيف تبني وكيلًا صوتيًا يتكلم باللهجة الخليجية على منصة Vapi](https://nutq.dev/ar/blog/arabic-voice-agent-vapi.md): أغلب منصات الوكلاء الصوتيين تتعثر حين يتكلم المتصل باللهجة الخليجية. كتلتان من إعدادات Vapi تصلحان طرفي المكالمة: سماع المتصل والرد عليه.
- [تحويل النص العربي إلى كلام في n8n: اجعل أي سير عمل يتكلم بصوت خليجي](https://nutq.dev/ar/blog/arabic-text-to-speech-n8n.md): لا عقدة إضافية تثبّتها ولا خطوة برمجية. عقدة HTTP Request واحدة وبيانات اعتماد واحدة، ويصبح كل سير عمل في n8n قادرًا على الكلام بالعربية — بصوت خليجي لا بصوت مذيع.
- [تحويل الكلام العربي إلى نص عبر API: دليل المطوّر لتفريغ اللهجة الخليجية](https://nutq.dev/ar/blog/arabic-speech-to-text-api.md): طلب POST واحد، وملف صوتي، ونص عربي يعود إليك واللهجة كما هي. ماذا ترسل، وماذا يعود، وكم يكلّف، وأين يخطئ التعرّف على الكلام العربي عادةً.
- [اللهجة الخليجية أم الفصحى: لماذا تعيد أغلب نماذج الكلام صياغة ما قاله عملاؤك؟](https://nutq.dev/ar/blog/gulf-arabic-vs-msa-transcription.md): المتصل يقول «شحالك» والنص يكتب «كيف حالك». يبدو تصحيحًا، لكنه تغيير في السجل — وأثره على البحث والتحليلات والامتثال أكبر مما يبدو.
- [كيف تفرّغ المكالمات الهاتفية العربية بدقة (ولماذا تفشل فيها أغلب النماذج)](https://nutq.dev/ar/blog/transcribe-arabic-phone-calls.md): خط الهاتف يرمي معظم الصوت الذي يصغي إليه نموذج الكلام. هذا ما يفعله ذلك بالتفريغ العربي، وما قسناه، وكيف تبني تفريغًا للمكالمات يصمد.
- [تحويل النص العربي إلى كلام مع استنساخ الصوت: دليل النطق الآلي باللهجة الخليجية](https://nutq.dev/ar/blog/arabic-text-to-speech-voice-cloning.md): ست إلى ثماني ثوانٍ من كلام شخص ما، والكلمات التي قالها، والنص الذي تريد قراءته — هذا كل ما يحتاجه استنساخ صوت خليجي. كيف تحصل على نتيجة طبيعية، والحيلة الواحدة التي تصلح أغلب أخطاء النطق.
- [كيف تضيف ترجمة عربية إلى الفيديو تلقائيًا (واللهجة كما هي)](https://nutq.dev/ar/blog/auto-arabic-subtitles-for-video.md): ارفع الفيديو، واضغط زرًا واحدًا، واحصل على ترجمة عربية مضبوطة على توقيت الكلام — باللهجة التي قيلت فعلًا. ثم نسّقها أو ترجمها أو ادمجها في الفيديو أو صدّرها ملف SRT.

## Optional

- [Blog categories](https://nutq.dev/blog): [Speech to text](https://nutq.dev/blog/category/speech-to-text), [Text to speech](https://nutq.dev/blog/category/text-to-speech), [Arabic dialects](https://nutq.dev/blog/category/arabic-dialects), [Video captions](https://nutq.dev/blog/category/video-captions), [On-premise](https://nutq.dev/blog/category/on-premise), [Guides](https://nutq.dev/blog/category/guides)
- [RSS (English)](https://nutq.dev/blog/rss.xml), [RSS (العربية)](https://nutq.dev/ar/blog/rss.xml)