This study investigates whether traditional denoising improves deep learning for Audio Event Detection (AED) in real-world, noisy environments like livestock farms. We evaluated three denoising algorithms — Spectral Subtraction, Adaptive Kalman Filter, and SD-ROM — on the aSwine dataset using state-of-the-art models, including Pretrained Audio Neural Networks (PANNs) and the Audio Spectrogram Transformer (AST). Contrary to conventional wisdom, all denoising methods proved detrimental, with a worst-case scenario showing a 65.8% mean Average Precision (mAP) decrease. We conclude that powerful models learn to be inherently noise-robust, making robust architectures a superior strategy to noise preprocessing.

Borehole image logs are critical for characterizing reservoirs in the petroleum industry. Exploring this data is challenging because image logs are extensive in size and demand multiple measurements to be juxtaposed side-by-side, such as resistivity and acoustic data. Despite its importance, no open-source applications exist to analyze such data. In this demonstration, we present ImageLogViewer, a Python-based tool for visualizing image logs. It can simultaneously display acoustic and dynamic/static micro-resistivity data with multiple display modes and custom windowing to facilitate the analysis of structures, e.g., fractures and cavities. It also integrates image processing, machine, and deep learning models for classifying and segmenting regions of interest.

From-scratch audio classification on weakly labeled data is essential for resource-constrained and domain-specific settings where large pretrained models are unavailable or prohibitively costly. We adopt multi-branch feature fusion as the core approach and compare CNN, Inception, Transformer, and Mamba architectures on the AudioSet balanced training split. Multi-branch fusion (per-feature encoders, merged before the head) improves all four architectures by 17–33% mAP. Five-fold CV shows Inception significantly outperforms the Transformer; Transformer and Mamba reach comparable multi-branch performance (p = 0.19), with preliminary results suggesting feature fusion may be at least as important as sequence model choice in this regime.