Mozilla Low-Resource ASR Challenge
Using high resource genetically related families to bolster ASR of low resource languages.
This project is a team submission for the Mozilla CommonVoice Spontaneous Speech ASR Challenge. We finetune Whisper for each language and also train adapters or task vectors (depending on the model) on genetically similar higher resource languages in order to bolster the performance on the low resource target languages.
This project is in progress.
Code for this project can be found here.
This project was done in collaboration with Dan DeGenaro and Xiulin Yang.