Text this: Improving Audio Recognition With Randomized Area Ratio Patch Masking: A Data Augmentation Perspective