This paper proposes two novel deep convolutional neural networks (CNN), called sparse coding convolutional neural network (SC-CNN) and multi-channel SC-CNN (MSC-CNN), to address the problem of sound event recognition and retrieval task. Unlike the general framework of a CNN, in which feature learning process is performed hierarchically, the proposed framework models the whole memorizing procedures in the human brain, including encoding, storage, and recollection. In particular, MSC-CNN is designed for recognition of multiple sound events which occur simultaneously. The experimental results indicated that the proposed SC-CNN and MSC-CNN outperformed the state-of-the-art systems in sound event recognition and retrieval.

Music information retrieval is of great interest in audio signal processing. However, relatively little attention has been paid to the playing techniques of musical instruments. This work proposes an automatic system for classifying guitar playing techniques (GPTs). Automatic classification for GPTs is challenging because some playing techniques differ only slightly from others. This work presents a new framework for GPT classification: it uses a new feature extraction method based on spectral-temporal receptive fields (STRFs) to extract features from guitar sounds. This work applies a supervised deep learning approach to classify GPTs. Specifically, a new deep learning model, called the hierarchical cascade deep belief network (HCDBN), is proposed to perform automatic GPT classification. Several simulations were performed and the datasets of: 1) data on onsets of signals; 2) complete audio signals; and 3) audio signals in a real-world environment are adopted to compare the performance. The proposed system improves upon the F-score by approximately 11.47% in setup 1) and yields an F-score of 96.82% in setup 2). The results in setup 3) demonstrate that the proposed system also works well in a real-world environment. These results show that the proposed system is robust and has very high accuracy in automatic GPT classification.

In the digital age, sharing moments through photos has become a daily habit. However, every face captured in these photos is vulnerable to unauthorized identification and potential misuse through AI-powered synthetic content generation. Previously, we introduced SnapSafe, a secure system for enabling selective image privacy focusing on facial regions for single-party scenarios. Recognizing that group photos with multiple subjects are a more common scenario, we extend SnapSafe to support multi-user facial privacy protection with dynamic access control designed for online photo platforms. Our approach introduces key splitting for access control, an owner-centric permission system for granting and revoking access to facial regions, and a request-based mechanism allowing subjects to initiate access permissions. These features ensure that facial regions remain protected while maintaining the visibility of non-facial content for general viewing. To ensure reproducibility and isolation, we implemented our solution using Docker containers. Our experimental assessment covered diverse scenarios, categorized as “Single”, “Small”, “Medium”, and “Large”, based on the number of faces in the photos. The results demonstrate the system’s effectiveness across all test scenarios, consistently performing face encryption operations in under 350 ms and achieving average face decryption times below 286 ms across various group sizes. The key-splitting operations maintained a $100\%$ success rate across all group configurations, while revocation operations were executed efficiently with server processing times remaining under 16 ms. These results validate the system’s capability in managing facial privacy while maintaining practical usability in online photo sharing contexts.

Faces and voices in live remote interactions are often treated as identity evidence, but advances in synthetic media generation weaken that inference. An attacker needs one feasible route to an incorrect identity or action decision; a defense must cover every route its policy admits under the stated attacker budget. This survey examines that generation–detection asymmetry across media checks, identity decisions, and action approval through five defense families grouped by evidence source and attacker influence. The first three families rely on attacker-influenced media: passive classification assesses caller-selected media, while behavioral challenges and physical probes constrain responses. The fourth family, trust infrastructure, constrains media entry and supports verifiable claims tied to systems or credentials. Out-of-band corroboration forms the fifth family and supplies evidence through a separately trusted route. The reviewed literature evaluates individual controls across all five families but does not establish complete-session reliability in live remote interactions. Audiovisual appearance alone cannot establish liveness, freshness, session binding, or identity correspondence. A convincing likeness should not be treated as authorization. Consequential actions require independently authenticated approval of the exact request through a separately trusted route. Finally, complete-session reliability should be assessed in realistic live sessions under adaptive attack.