PEFT for Speech: Unveiling Optimal Placement, Merging Strategies, and Ensemble Techniques

• Published in: 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) pp. 705 - 709
• Main Authors: Lin, Tzu-Han, Wang, How-Shing, Weng, Hao-Yung, Peng, Kuang-Chen, Chen, Zih-Ching, Lee, Hung-Yi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.04.2024
• Subjects: Adaptation models; adapters; Conferences; ensemble learning; Merging; network architecture search; Parameter-efficient learning; Self-supervised learning; Signal processing; Speech enhancement; Speech recognition
• DOI: 10.1109/ICASSPW62465.2024.10625957

Parameter-Efficient Fine-Tuning (PEFT) is increasingly recognized as an effective method in speech processing. However, the optimal approach and the placement of PEFT methods remain inconclusive. Our study conducts extensive experiments to compare different PEFT methods and their layer-wise placement adapting Differentiable Architecture Search (DARTS). We also explore the use of ensemble learning to leverage diverse PEFT strategies. The results reveal that DARTS does not outperform the baseline approach, which involves inserting the same PEFT method into all layers of a Self-Supervised Learning (SSL) model. In contrast, an ensemble learning approach, particularly one employing majority voting, demonstrates superior performance. Our statistical evidence indicates that different PEFT methods learn in varied ways. This variation might explain why the synergistic integration of various PEFT methods through ensemble learning can harness their unique learning capabilities more effectively compared to individual layer-wise optimization.