Yanrice/lingala-asr

# Lingala ASR Dataset An open-source Automatic Speech Recognition (ASR) dataset for Lingala, a Bantu language spoken by over 10 million people in the Democratic Republic of Congo and Republic of Congo. ## 📊 Dataset Overview - **Language**: Lingala (Kinshasa dialect) - **Total Recordings**: 870+ audio clips - **Audio Format**: WAV, 16kHz, mono - **Duration**: ~5 seconds per clip - **Total Duration**: ~72 minutes - **Dialect**: Kinshasa variant (using "naza" instead of "nazali", "eza" instead of "ezali") ## 🎯 Purpose This dataset was created to train speech recognition models for Lingala, an underrepresented language in NLP and ASR research. The goal is to enable: - Automatic transcription of Lingala speech - Translation of Lingala audio to other languages - Captioning for Lingala YouTube videos and media content - Accessibility tools for Lingala speakers ## 📁 Dataset Structure ``` lingala-asr/ ├── data/ │ ├── audio/ │ │ ├── clip0001.wav │ │ ├── clip0002.wav │ │ └── ... │ └── transcriptions.txt └── README.md ``` ### Transcriptions Format The `transcriptions.txt` file contains one entry per line in the format: ``` filename.wav|Lingala transcription text ``` Example: ``` clip0001.wav|Mbote clip0002.wav|Naza malamu clip0003.wav|Sango nini? ``` ## 🗣️ Content Categories The dataset includes diverse phrases across multiple domains: - **Greetings & Introductions**: Basic conversational phrases - **Daily Activities**: Common actions and routines - **Food & Drink**: Kitchen, cooking, and dining - **Family & Relationships**: Family members and social connections - **Locations & Directions**: Navigation and places - **Time & Dates**: Temporal expressions - **Emotions & Feelings**: Expressing sentiments - **Health & Medical**: Medical terminology and health-related phrases - **Shopping & Market**: Commercial transactions - **Technology**: Modern tech vocabulary - **Transportation**: Vehicles and travel - **Work & Business**: Professional contexts - * …