PosCUDA: Position based Convolution for Unlearnable Audio Datasets
Vignesh Gokul, Shlomo Dubnov
TL;DR
PosCUDA addresses the privacy risk of unauthorized audio model training by creating unlearnable datasets through class-specific positional blurs applied to small audio patches. It uses private-key driven 1-D filters to blur localized regions, embedding a label-specific mapping that impedes generalization while preserving overall audio quality. Empirical results show strong unlearnability across multiple architectures (CNN, LSTM, Transformer) and input representations, with minimal degradation measured by Fréchet Audio Distance. The approach offers practical protection for audio content and points to future work extending to longer audio, unsupervised learning, and generative models.
Abstract
Deep learning models require large amounts of clean data to acheive good performance. To avoid the cost of expensive data acquisition, researchers use the abundant data available on the internet. This raises significant privacy concerns on the potential misuse of personal data for model training without authorisation. Recent works such as CUDA propose solutions to this problem by adding class-wise blurs to make datasets unlearnable, i.e a model can never use the acquired dataset for learning. However these methods often reduce the quality of the data making it useless for practical applications. We introduce PosCUDA, a position based convolution for creating unlearnable audio datasets. PosCUDA uses class-wise convolutions on small patches of audio. The location of the patches are based on a private key for each class, hence the model learns the relations between positional blurs and labels, while failing to generalize. We empirically show that PosCUDA can achieve unlearnability while maintaining the quality of the original audio datasets. Our proposed method is also robust to different audio feature representations such as MFCC, raw audio and different architectures such as transformers, convolutional networks etc.