lberniilyas/DatasetRecorder

A Python web app for building a Darija speech dataset — runs on PC but streams to any device over a Cloudflare Tunnel, so you can record straight from your iPhone's high-quality mic instead of being stuck with a mediocre laptop microphone. Sentences are served one by one and audio is saved automatically to the PC. # BLANI Darija Recorder A small mobile-first web tool to record Darija audio for fine-tuning your Whisper-Darija model. You see a sentence, tap record, speak, tap stop, save. Audio + transcript metadata get written to disk as a HuggingFace-ready dataset. ## What you'll need - Python 3.10+ - `ffmpeg` installed and on your PATH (you already have it from the BLANI backend work) - A `GEMINI_API_KEY` (only used **once**, to generate the script list) - An **ngrok** free account, if you want to record from your iPhone ## Setup ```bash # from this folder pip install -r requirements.txt cp .env.example .env # edit .env, paste your GEMINI_API_KEY ``` ## Step 1 — Generate the script list This calls Gemini once to produce 30 Darija sentences (focused on hotels, apartments, and trip planning per your scope), in Arabic script + Latin transliteration + English translation. Writes `scripts.json` in the same folder. ```bash python generate_scripts.py ``` You can re-run this any time to refresh the script set. The server reads `scripts.json` on every request, so changes are picked up live. **Review `scripts.json` before recording.** If any sentence looks weird or wrong, edit the file by hand — the file is plain JSON. ## Step 2 — Start the server ```bash python server.py ``` You'll see: ``` open: localhost ngrok: ngrok http 5050 (for iPhone access) ``` Open `localhost` in your browser to test on your PC. Allow microphone access when prompted. ## Step 3 — Record from your iPhone (HTTPS via ngrok) iPhone Safari requires HTTPS to use the microphone. `localhost` doesn't qualify — you need a public HTTPS URL pointing at your PC. ngrok provides this for free. 1. Sign up at ngrok.com (free tier is fine) 2. Install ngrok, run `ngrok config add-authtoken ` once 3. In a **second terminal**, while `server.py` is running: ```bash ngrok http 5050 ``` 4. ngrok prints something like `Forwarding abc123.ngrok-free.app -> localho …