kingabzpro/WOLOF-ASR-Wav2Vec2

Audio Preprocessing and finetuning of wav2vec2-large-xlsr model on AI4D Baamtu Datamation - Automatic Speech Recognition in WOLOF Data. # WOLOF-ASR-Wav2Vec2 Audio preprocessing and finetuning using wav2vec2-large-xlsr model on AI4D Baamtu Datamation - automatic speech recognition in WOLOF data. ## Zindi ASR Competition The challenge will focus on a public transport use case for two reasons. First, many users of public transport can’t read or speak French, so they can’t interact with existing apps that help passengers to find a bus for a given destination. And second, there is already an existing app in Senegal, WeeGo, which helps passengers to get transport information. The goal of this competition is to build an ASR model that will help illiterate people use existing apps to find which bus they can take to reach their destination, without having to know how to read or write. **About Baamtu Datamation (****baamtu.com****)** Baamtu Datamation is a Senegalese company focused on helping companies to leverage AI and Big Data. **About AI4D-Africa; Artificial Intelligence for Development-Africa Network (****ai4d.ai****)** ## WOLOF Speech Data The Zindi data is easily accessible if you have Zindia account. **I don't have permission to publish the data in repos due to their updates in terms and conditions.** There are 6683 audio files in the train set and 1590 in the test set. You will use these files to train your model and submit your translations. The goal of this competition is to build an ASR model that will help illiterate people use existing apps to find which bus they can take to reach their destination, without having to know how to read or write. **Files available for download:** - **clips.zip** - contains all the audio files - **Train.csv** - contains the audio IDs and the transcription. This is the dataset that you will use to train your model. - **Test.csv** - contains the audio IDs and some information about the audio file. It does not include the transcription. You will use the model you trained on Train.csv to make your translation predictions for the test files. - **S …