Investigating the Sensitivity of Pre-trained Audio Embeddings to Common Effects