Adding audio samples(non supervised) to XLSR-53 model

[arikhalperin]

I have already used XLSR-53 successfully, fine-tuning it according : https://huggingface.co/blog/fine-tune-xlsr-wav2vec2

However I have a lot of unlabeled data. I understand that building a Wav2Vec model from scratch does not make sense, so I want to train a model With my audio but use XLSR-53 as a base - but I could not find an explanation how to do it.