Design of a Distributed Music Copyright Protection Model Using Federated Learning

Abstract

The increasing severity of music copyright infringement exposes the limitations of centralized protection systems, especially in data privacy, cross-platform collaboration, and single-point failure risk. This paper proposes a distributed music copyright protection model based on federated learning. Because audio fingerprinting, encrypted gradient aggregation, and copyright verification all rely on robust signal representation in networked digital systems, the method also has engineering relevance for secure signal-processing applications. The proposed model adopts a three-layer architecture: the client layer extracts music fingerprint features using Mel-frequency cepstral coefficients (MFCCs), the aggregation layer combines model parameters from participating nodes through secure multi-party computation, and the blockchain layer stores copyright information in a decentralized manner. In implementation, a convolutional neural network copyright recognition model is trained locally by each participant; encrypted gradients are uploaded to the coordinating server; federated averaging updates the global model until convergence; and copyright-related records are stored through blockchain mechanisms. Experiments on 50,000 licensed music tracks show that the proposed model improves accuracy by 2.1 percentage points and recall by 2.3 percentage points compared with centralized methods. Communication overhead is reduced by 79.4%, while data independence and effective copyright protection are maintained across participating platforms.