The objective of this challenge is to create an automatic speech recognition model on Luganda. You will train your models on the complete Luganda dataset provided by Mozilla Common Voice and you will test your model on the ~7 000 in the test set provided.
Please note that you must use machine learning in your solution and any use of data leaks will result in your disqualification. We reserve the right to review anyone's code at any time during and after the close of this challenge.
You may use the Kinyarwanda dataset on Mozilla Common Voice to create a pre-trained model for transfer learning. You can use any other open-source pre-trained models, but not additional datasets.
Files available for download:
First, go to Mozilla Common Voice and download the Luganda dataset here to train your models on. When you have done that, you can download the audio test set provided here on Zindi. Do not download the audio test set here until you have download the Luganda dataset - this is a requirement for the competition!
Test.csv - contains the IDs of the audio files that you will test your model on and some macro data on each of the audio files.
SampleSubmission.csv - shows the submission format for this competition, with the ‘Audio_ID’ column mirroring that of Test.csv and the ‘Target’ column containing your predictions. The order of the rows does not matter, but the names of the 'Audio_ID’ must be correct.
test_audio.zip - contains all the audio files for this challenge.