Efficient End-to-End Noise-Invariant Speech Features via Multi-Task Learning

RobustDistiller enhances noise-invariant speech features through multi-task learning, improving model robustness and compression efficiency.