Information Retrieval in Digital Sound Archives and Libraries: Preservation of the Documents of Xinjiang-Style Guzheng Music through Digital Library Curation

Abstract

The preservation and accessibility of cultural artefacts are increasingly reliant on the integration of technology into archival practices. Digital libraries play an essential role in safeguarding musical heritage, offering scalable storage, improved accessibility, and long term sustainability. Sound archives face degradation due to inadequate organisational systems and limited public usage. To address this, the proposed Information Retrieval Framework (IRF) aims to establish a new system for the protection of Guzheng musical heritage in Xinjiang by advancing audio curation techniques. This framework employs a Hybrid Cascaded Convolutional Gated Neural Network (HCC-GNN), which integrates Cascaded Convolutional Neural Networks (CCNNs) to extract robust short-term features from raw waveforms and spectrograms. These features are subsequently processed through a Gated Recurrent Unit (GRU) and Recurrent Neural Network (RNN) to capture long-term temporal patterns in music recordings. This enables the effective tagging and retrieval of archived audio content, including Xinjiang-style Guzheng performances, even amidst background noise and repeated acoustic events. An evaluation was carried out using Xinjiang-style Guzheng recordings from digital archives. Prior sound normalisation through Wiener filtering enhanced audio clarity, while Wavelet Transform and data augmentation optimised the extraction process, increasing diversity and improving system reliability. A retrieval-optimised network then processed these refined features, enabling users to swiftly obtain relevant information from extensive digital audio databases. The HCC-GNN model outperformed conventional methods, achieving an accuracy of 97.5%, Precision@10 of 97%, Recall@10 of 95.8%, Mean Average Precision (MAP) of 96%, and an F1 Score of 95.2%. These results promote cultural sustainability and enhance accessibility for scholars