Search

It is a well-established practice to build a robust system for sound event detection by training supervised deep learning models on large datasets, but audio data collection and labeling are often challenging and require large amounts of effort. This paper proposes a workflow based on few-shot metric learning for emergency siren detection performed in steps: prototypical networks are trained on publicly available sources or synthetic data in multiple combinations, and at inference time, the best knowledge learned in associating a sound with its class representation is transferred to identify ambulance sirens, given only a few instances for the prototype computation. Performance is evaluated on siren recordings acquired by sensors inside and outside the cabin of an equipped car, investigating the contribution of filtering techniques for background noise reduction. The results show the effectiveness of the proposed approach, achieving AUPRC scores equal to 0.86 and 0.91 in unfiltered and filtered conditions, respectively, outperforming a convolutional baseline model with and without fine-tuning for domain adaptation. Extensive experiments conducted on several recording sensor placements prove that few-shot learning is a reliable technique even in real-world scenarios and gives valuable insights for developing an in-car emergency vehicle detection system. [ABSTRACT FROM AUTHOR]

In the fields of audio analysis, processing and acoustic modelling, Deep Learning has been swiftly adopted, initially borrowing their methods from the image processing and computer vision field, and then finding creative and innovative solutions to suit domain-specific needs of acoustic research. In Navarro-Caceres et al. [[8]] the authors propose an assistive music composition system that is able to satisfy multiple criteria and find multiple candidates in chord sequence generation using an Artificial Immune System, a biologically inspired AI technique based on Genetic Programming for multi-objective optimisation. The generated chord progressions fit a desired tonal tension profile, taking for instance consonance or the melodic attraction of successive chords into account, thus aiding the composition process. [Extracted from the article]

Audio equalization is an active research topic aiming at improving the audio quality of a loudspeaker system by correcting the overall frequency response using linear filters. The estimation of their coefficients is not an easy task, especially in binaural and multipoint scenarios, due to the contribution of multiple impulse responses to each listening point. This paper presents a deep learning approach for tuning filter coefficients employing three different neural networks architectures—the Multilayer Perceptron, the Convolutional Neural Network, and the Convolutional Autoencoder. Suitable loss functions are proposed for each architecture, and are formulated in terms of spectral Euclidean distance. The experiments were conducted in the automotive scenario, considering several loudspeakers and microphones. The obtained results show that deep learning techniques give superior performance compared to baseline methods, achieving almost flat magnitude frequency response. [ABSTRACT FROM AUTHOR]

In the last few years, due to the technological improvement of advanced metering infrastructures, water and natural gas grids can be regarded as smart-grids, similarly to power ones. However, considering the number of studies related to the application of computational intelligence to distribution grids, the gap between power grids and water/gas grids is notably wide. For this purpose, in this paper, a framework for leakage identification is presented. The framework is composed of three sections aimed at the extraction and the selection of features and at the detection of leakages. A variation of the Sequential Feature Selection (SFS) algorithm is used to select the best performing features within a set, including, also, innovative temporal ones. The leakage identification is based on novelty detection and exploits the characterization of a normality model. Three statistical approaches, The Gaussian Mixture Model (GMM), Hidden Markov Model (HMM) and One-Class Support Vector Machine (OC-SVM), are adopted, under a comparative perspective. Both residential and office building environments are investigated by means of two datasets. One is the Almanac of Minutely Power dataset (AMPds), and it provides water and gas data consumption at 1, 10 and 30 min of time resolution; the other is the Department of International Development (DFID) dataset, and it provides water and gas data consumption at 30 min of time resolution. The achieved performance, computed by means of the Area Under the Curve (AUC), reaches 90% in the office building case study, thus confirming the suitability of the proposed approach for applications in smart water and gas grids. [ABSTRACT FROM AUTHOR]