Arabic speech to text
Arabic speech to text that understands Saudi dialect.
Accurate Saudi and Gulf Arabic transcription, live as people speak or from recordings up to two hours long.
Speech to text
Live text streams here while you speak.
Voice
Model
Nabrah TTS v1
At a glance
- Dialects
- Saudi and Gulf Arabic
- Live
- Real-time WebSocket with partial and final results
- Files
- Async jobs for recordings up to two hours
- Accuracy
- Priority words for names and terms
Transcription that keeps up with the region.
- 01
Saudi and Gulf Arabic
Recognition tuned to how people in the region actually speak.
- 02
Real-time results
Partial text while people speak, and a final segment when they pause.
- 03
Long recordings
Async jobs for files up to two hours.
- 04
Priority words
Bias recognition toward the names and terms that matter to you.
- 05
Silence detection
Choose how much silence ends an utterance on live audio.
- 06
Studio and API
Transcribe in the browser with Studio, or from your servers through the API.
Real time
Text as fast as people talk.
Open the speech-to-text WebSocket, send raw PCM audio (16-bit, 16 kHz, mono), and receive partial transcripts while someone speaks, then a final segment when they pause.
Open the socket from your server: the API key travels in its first message.
import WebSocket from "ws";
// Open the socket from your server: the key travels in the first message.
const ws = new WebSocket("wss://api.nabrah.ai/api/ext/stt/ws");
ws.on("open", () => {
ws.send(JSON.stringify({ api_key: process.env.NABRAH_API_KEY, priority_words: ["نبرة"] }));
});
ws.on("message", (data) => {
const msg = JSON.parse(data);
if (msg.type === "ready") mic.on("data", (pcm) => ws.send(pcm)); // 16-bit, 16 kHz, mono
if (msg.type === "transcript") console.log(msg.is_final ? "final" : "partial", msg.text);
});{ "type": "transcript", "text": "حاب أحجز موعد بكرة", "is_final": true }Trimmed for clarity. Every field is in the API reference.
Recorded audio
Hours of audio, one request.
Upload a call recording, a meeting or a voice note. It runs as a job: check its status, then read the transcript. Recordings can be up to two hours long.
Transcription jobs
Recordings up to two hours, transcribed in the background.
- call-recording.mp304:12Processing
- weekly-meeting.wav47:30Completed
- voice-note.mp300:38Queued
Priority words
Transcript
Transcribing…
How it works
From audio to text in three steps.
Live or recorded, the flow is the same.
- 1
Send the audio
Speak into Studio, upload a recording, or stream 16 kHz PCM to the real-time socket.
- 2
Nabrah transcribes it
Live audio returns partial text as people speak and a final segment when they pause. Recordings run as jobs.
- 3
Use the text
Copy it from Studio, read the job result or the transcript events, and feed it into your own workflows.
Where teams use Nabrah transcription.
Anywhere spoken Arabic needs to become text you can search and act on.
Transcribe calls for quality review, search and analytics.
Caller
Contact centers
Frequently asked questions
Straight answers about Nabrah Speech to Text. Anything else, our team is a message away.
Nabrah Speech to Text (STT) converts spoken Arabic into text. It is built for Saudi and Gulf dialects and works live over a WebSocket or on recorded files as async jobs, in Nabrah Studio or through the API.
It is built for Saudi and Gulf (Khaleeji) Arabic, the way customers in the region actually speak, rather than Modern Standard Arabic only.
Yes. Stream raw PCM (16-bit, 16 kHz, mono) to the real-time WebSocket and receive partial transcripts as people speak and a final segment when they pause.
Async transcription accepts recordings up to two hours long. Upload the file, then check the job until the transcript is ready.
Yes. Pass priority words, such as product names, places or people, and set how strongly recognition should favour them.
It uses credits from your plan: 100 credits per minute of audio. You can start with 7,000 free credits, no credit card required.