Luganda Voice Commands Dataset

The dataset contains audio files of voice commands in the Luganda language. A total of 10,200 audio files were collected from 81 participants (43 males and 38 females, aged 20 to 25), divided into two sets: 9,132 for training and 1,068 for testing. Special attention was paid to the test set, ensuring that audio files used for this set were recorded from 7 (4 male and 3 female) speakers, and none appeared in the training set. This was done to simulate real-world scenarios where the model may face unseen speakers during deployment, thus assessing the model’s robustness. This dataset can be used for SLU tasks like speech intent classification and other NLP tasks.