unsupervised speech recognition has shown great potential to make automatic speech recognition (ASR) systems accessible to every language. However, existing methods still heavily rely on hand-crafted pre-processi