Efficient audio deepfake detection with smaller models
Task-Aware Joint Pruning and Distillation for Efficient Audio Deepfake Detection
SoundArtificial Intelligence
Summary
Deepfake voices are getting better and more convincing, which can cause security problems. The authors focus on making deepfake speech detection work well on small devices by shrinking the size of the detection model without losing much accuracy. They created a new method that smartly removes unnecessary parts of the model and transfers important knowledge from bigger models to smaller ones. Their smaller model keeps most of the accuracy while being much faster and easier to run on devices like phones.
What this means in practice
- •For mobile app developers: Deploy accurate deepfake speech detectors that run efficiently on smartphones and other limited hardware.
- •For security system integrators: Build cost-effective audio authentication tools that fit within tight computational budgets while maintaining detection reliability.
Authors
Miao He, Peng Cheng, Zhongjie Ba, Qing Wen, Li Lu, Xin Yang, Kui Ren
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.