The spreading of urban areas and the growth of human population worldwide raise societal and environmental concerns. To better address these concerns, the monitoring of the acoustic environment in urban as well as rural or wilderness areas is an important matter. Building on the recent development of low cost hardware acoustic sensors, we propose in this paper to consider a sensor grid approach to tackle this issue. In this kind of approach, the crucial question is the nature of the data that are transmitted from the sensors to the processing and archival servers. To this end, we propose an efficient audio coding scheme based on third octave band spectral representation that allows: (1) the estimation of standard acoustic indicators; and (2) the recognition of acoustic events at state-of-the-art performance rate. The former is useful to provide quantitative information about the acoustic environment, while the latter is useful to gather qualitative information and build perceptually motivated indicators using for example the emergence of a given sound source. The coding scheme is also demonstrated to transmit spectrally encoded data that, reverted to the time domain using state-of-the-art techniques, are not intelligible, thus protecting the privacy of citizens.

The impact of urban sound on human beings has often been studied from a negative point of view (noise pollution). In the two last decades, the interest of studying its positive impact has been revealed with the soundscape approach (resourcing spaces). The literature shows that the recognition of sources plays a great role in the way humans are affected by sound environments. There is thus a need for characterizing urban acoustic environments not only with sound pressure measurements but also with source-specific attributes such as their perceived time of presence, dominance or volume. This paper demonstrates, on a controlled dataset, that machine learning techniques based on state of the art neural architectures can predict the perceived time of presence of several sound sources at a sufficient accuracy. To validate this assertion, a corpus of simulated sound scenes is first designed. Perceptual attributes corresponding to those stimuli are gathered through a listening experiment. From the contributions of the individual sound sources available for the simulated corpus, a physical indicator approximating the perceived time of presence of sources is computed and used to train and evaluate a multi-label source detection model. This model predicts the presence of simultaneously active sources from fast third octave spectra, allowing the estimation of perceptual attributes such as pleasantness in urban sound environments at a sufficient degree of precision.

Machine listening systems for environmental acoustic monitoring face a shortage of expert annotations to be used as training data. To circumvent this issue, the emerging paradigm of self-supervised learning proposes to pre-train audio classifiers on a task whose ground truth is trivially available. Alternatively, training set synthesis consists in annotating a small corpus of acoustic events of interest, which are then automatically mixed at random to form a larger corpus of polyphonic scenes. Prior studies have considered these two paradigms in isolation but rarely ever in conjunction. Furthermore, the impact of data curation in training set synthesis remains unclear. To fill this gap in research, this article proposes a two-stage approach. In the self-supervised stage, we formulate a pretext task (Audio2Vec skip-gram inpainting) on unlabeled spectrograms from an acoustic sensor network. Then, in the supervised stage, we formulate a downstream task of multilabel urban sound classification on synthetic scenes. We find that training set synthesis benefits overall performance more than self-supervised learning. Interestingly, the geographical origin of the acoustic events in training set synthesis appears to have a decisive impact.

Bandwidth extension has a long history in audio processing. While speech processing tools do not rely on side information, production-ready bandwidth extension tools of general audio signals rely on side information that has to be transmitted alongside the bitstream of the low frequency part, mostly because polyphonic music has a more complex and less predictable spectral structure than speech.This paper studies the benefit of considering a dilated fully convolutional neural network to perform the bandwidth extension of musical audio signals with no side information on the magnitude spectra. Experimental evaluation using two public datasets, medley-solos-db and gtzan, respectively of monophonic and polyphonic music demonstrate that the proposed architecture achieves state of the art performance.Index Terms-Artificial audio bandwidth extension, deep neural network, musical audio processing

As part of the Agence Nationale de Recherche Caractérisation des ENvironnements SonorEs urbains (Characterization of urban sound environments) project, a questionnaire was sent in January 2019 to households in a 1 km 2 study area in the city of Lorient, France, to which about 318 responded. The main objective of this questionnaire was to collect information about the inhabitants' perception of the sound environments in their neighborhoods, streets, and dwellings. In the same study area, starting mid-2019, about 70 sensors were continuously positioned, and 15 of them were selected for testing sound source recognition models. The French lockdown due to the COVID-19 crisis occurred during the project, and the opportunity was taken to send a second questionnaire during April 2020. About 31 of the first 318 first survey respondents answered this second questionnaire. This unique longitudinal dataset, both physical and perceptual, allows the undertaking of an analysis from different perspectives of such a period. The analysis reveals the importance of integrating source recognition tools, soundscape observation protocol, in addition to physical level analysis, to accurately describe the changes in the sound environment.