Using Radio Archives for Low-Resource Speech Recognition

Unsupervised speech representation learning from radio archives to faciliate speech recognition for Maninkakan, Pular and Susu, which are currently under-served by language technology. Datasets, Wav2vec encoder, and other resources for West African Speech technology development. Prototype of a virtual assistant for contact management in Maninkakan, Pular and Susu.
[code] [paper]