KabulaBenjamin/kabras-ai-project

# Kabras AI Translator A speech translation app for the **Kabras language** — a Luhya dialect spoken in Western Kenya. Built to support language preservation through accessible AI tooling. ## Features - 🎙️ Live microphone recording or audio file upload - 🔍 Exact and fuzzy matching against a growing Kabras lexicon - 📊 Confidence scoring with top alternative matches - 📝 Dual logging to JSON and CSV for research use - ➕ In-app lexicon expansion with duplicate detection - 🔎 Admin gap analysis — surfaces most-needed phrases by frequency - 📤 Export translation history as JSON or CSV ## Project structure ``` kabras-ai-project/ ├── app/ │ ├── gradio_app.py # Main Gradio UI and all logic │ ├── main.py # Entry point — run this │ ├── lexicon/ │ │ └── kabras_lexicon.json │ ├── datasets/ │ │ └── translations_log.csv # Auto-generated on first translation │ └── config/ │ └── settings.yaml ├── docs/ │ └── README.md └── requirements.txt ``` ## Setup ```bash git clone github.com cd kabras-ai-project python -m venv .venv .venv\Scripts\activate # Windows pip install -r requirements.txt ``` ## Run ```bash python app/main.py ``` Then open 127.0.0.1 in your browser. ## How it works 1. Audio is captured via microphone or file upload 2. Google Speech Recognition transcribes it (Swahili first, English fallback) 3. The transcription is matched against `kabras_lexicon.json`: - **Exact match** → 100% confidence - **Fuzzy match** (≥ 62% similarity) → confidence score + alternatives shown - **No match** → logged and flagged for lexicon expansion 4. Every attempt is logged to `datasets/translations_log.csv` ## Research companion This app generates the dataset used by the kabras-asr-research project for WER analysis and confidence visualisations. ## Contributing To add new Kabras phrases, use the **Expand Lexicon** tab in the app, or edit `lexicon/kabras_lexicon.json` direct …