jnlandu/lingala-stt

An effort to create, collect audio, and curate, and plush a high-quality lingala speech corpus for ASR. # Central African Languages Speech Corpus An effort to create, curate, and openly publish a high-quality Lingala speech corpus for automatic speech recognition (ASR) research and products. **Now with automated data collection!** GitHub Actions automatically downloads new Radio Okapi audio every 12 hours across multiple Congolese languages and stores them to Google Drive. --- ## Repository Structure ``` lingala-stt/ ├── README.md ← you are here ├── LICENSE ← CC-BY-4.0 ├── .gitignore ├── requirements.txt ← Python dependencies ├── data/ │ └── raw/ │ └── okapi/ ← automated Radio Okapi downloads │ ├── lingala/ ← Lingala audio bulletins │ │ ├── *.mp3 │ │ └── metadata/ │ ├── kikongo/ ← Kikongo audio bulletins │ │ ├── *.mp3 │ │ └── metadata/ │ ├── tshiluba/ ← Tshiluba audio bulletins │ │ ├── *.mp3 │ │ └── metadata/ │ └── manifest.json ← multi-language dataset catalog ├── scripts/ │ ├── download_okapi.py ← automated Radio Okapi scraper (multi-language) │ ├── schedule_okapi.py ← local scheduling script │ ├── align_whisper.py ← auto-transcribe + forced alignment │ └── segment.py ← silence-based segmentation ├── logs/ ← scraper logs and download reports ├── .github/ │ └── workflows/ │ ├── radio_okapi.yml ← 12h collection workflow │ └── radio_okapi_daily.yml ← daily collection workflow └── top-congo/ ← additional dataset sources ``` ## Audio Sources ### Automated Collection (Primary Source) ### Automated Collection (Primary Source) - **🤖 Radio Okapi**: Daily news bulletins automatically collected in multiple languages and stored to Google Drive - **Lingala**: `radiookapi.net` - **Kikongo**: `radiookapi.net` - **Tshiluba**: `rad …