Task-Aware Joint Pruning and Distillation for Efficient Audio Deepfake Detection

2026-10-04

AI Summary

This study addresses the challenge of deploying high-parameter self-supervised learning models for audio deepfake detection on resource-constrained devices by proposing a task-aware joint pruning and distillation framework. The method introduces the first forgery-discriminative joint optimization mechanism, overcoming the performance bottlenecks inherent in conventional content-centric compression approaches. Efficient model compression is achieved through the synergy of cross-domain knowledge distillation and motion-guided structured pruning. Experimental results demonstrate that the compressed model contains only 31.9M parameters with a 6.3-fold reduction in computational cost, while the average performance across multiple datasets degrades by merely 1.3%. By preserving strong generalization capabilities, the proposed approach exhibits significant potential for edge deployment.

Abstract

Advances in speech synthesis have made deepfake speeches increasingly convincing, posing growing threats to security. While self-supervised learning (SSL) based detectors achieve state-of-the-art performance, their computational demands (typically 300M+ parameters) prevent deployment on resource-constrained devices. Existing compression methods, designed mainly for content-centric tasks, struggle to maintain competitive performance when directly adapted to deepfake detection. We propose a Task-Aware Joint Pruning and Distillation framework that combines cross-domain knowledge distillation with movement-guided structured pruning to transfer forgery-discriminative knowledge and preserve critical structures under aggressive compression. Our framework reduces the model to 31.9M parameters with 6.3$\times$ FLOPs reduction, with an average performance drop of only 1.30\% across multiple datasets compared to the uncompressed baseline, demonstrating strong potential for on-device deployment.

Problem

• Audio Deepfake Detection
• Model Compression
• Self-Supervised Learning
• Resource-Constrained Devices

Innovation

• Audio Deepfake Detection
• Knowledge Distillation
• Structured Pruning
• Model Compression
• Self-Supervised Learning