Your Darija, written.
هضرتك، مكتوبة.
Drop a recording. Kteb writes what was said in Moroccan Darija. You fix the mistakes in seconds, and you leave with clean text or subtitles.
- ✓Runs on your computer. Nothing is uploaded.
- ✓Free and open source
- ✓Understands Darija mixed with French and English
- Step 1 of 4Choose or record
- Step 2 of 4Wait for the text
- Step 3 of 4Correct the text
- Step 4 of 4Export
Step 1 of 4Choose a file or record your voice
Choose one audio or video file that contains speech in Moroccan Darija. Common formats work: MP3, WAV, M4A, OGG, MP4, MKV, MOV. The file is read by the Kteb server on this computer. It is not sent to the internet.
or
Speak in Darija. Your browser first asks for permission to use the microphone. Nothing is recorded before you allow it, and the recording stops when you click Stop, or after 15 minutes.
Recording 0:00
Why this project existsAn open project for everyone who speaks Darija
مشروع مفتوح لكل واحد كيهضر بالدارجة.
Millions of people speak Darija every day, but very little of it is written down, and most software does not understand it. Kteb is free and its code is open, so that anyone can use it, check it and improve it. Here is who it can help:
- People who make videosSubtitles let more people follow a video, also with the sound off. Writing them by hand takes hours.
- Deaf and hard-of-hearing peopleA video in Darija with subtitles is a video they can follow.
- Students, journalists and researchersAn interview or a lesson becomes text that can be searched, quoted and kept.
- Darija itselfSpeech models learn from examples of speech with correct text. Each text you correct can become one of those examples, and help the next open model understand Darija better.
Your recordings stay on your computer. Nothing is shared unless you decide to share it.
Step 2 of 4Wait for the text
Step 3 of 4Correct the text
The model gets about half of the words wrong on average, so every part needs your eyes. Each word you fix is one more verified example of written Darija.
parts checked
How to correct (5 rules)
- Each row is one part of the recording. The button shows when the part starts and ends.
- Click the button to hear only that part. A part counts as checked (✓) when you have played it or clicked in its text.
- Click the text and type to correct it. A corrected row gets a red line and an Undo button. Undo brings back the text the model wrote.
- In a text box: Ctrl+Enter plays the part again. Enter goes to the next part and plays it. Shift+Enter adds a new line.
- Fastest way: click the first text, listen, fix, press Enter, repeat.
Step 4 of 4Export
Each button saves one file to your browser's download folder. The file name starts with the name of your recording. Your corrections exist only in this page until you export them.
Text with start and end times. For video editors, VLC, YouTube and most video tools.
The text only, one part per line, no times. For reading or pasting into a document.
The same subtitles in the format that web pages use.
Only the parts you changed: the text the model wrote and your corrected text. Keep it if you want to help improve the model later.