Lifelong audio feature extraction involves learning new sound classes
incrementally, which is essential for adapting to new data distributions over
time. However, optimizing the model only on new data can lead to catastrophic
forgetting of previously learned tasks, which undermines the