Increasing urbanisation poses new challenges in mitigating noise pollution and preserving quality of life. In this study, we present an innovative approach for the classification of environmental noise, exploiting advanced Deep Learning (DL) techniques. By merging three diferent public datasets, we created a unified corpus to train and test a convolutional neural network (CNN), with the aim of eficiently recognising and classifying various noise events. The proposed approach overcomes the limitations of conventional methodologies, avoiding the need for data pre-processing that could alter sound characteristics. The experimental results demonstrate a significant improvement in classification accuracy, reaching 96.93% with the test set and 100% by applying a post-processing filter. These results emphasise the potential of DL in the treatment of environmental noise, ofering new perspectives for signal processing and telecommunications.
The search for sustainable solutions to mitigate the impact
of environmental noise has become crucial to preserving
the quality of life in our increasingly urban society. In this
context, the recognition and classification of environmental
noise emerge as key challenges in the field of signal
processing and telecommunications, where noise can significantly
degrade the quality and intelligibility of transmitted signals
[
Pioneering studies, such as that of Couvreur et al. [
This study demonstrates how the application of DL techniques can ofer efective solutions to the challenges of environmental noise classification, with potential significant benefits for the telecommunication sector and society at large. Our research opens new perspectives for the use of artificial intelligence in urban noise mitigation, promoting a more sustainable environment and a better quality of life.
In section 2 we discuss the importance of developing effective noise classification strategies, which are essential for improving the quality of communication and, consequently, the quality of life in urban areas.
In section 3, we present our innovative approach, which exploits advanced DL techniques for analysing and classifying environmental noise. We will illustrate how, through the use of Convolutional Neural Networks (CNNs), our model works directly with the raw audio data, avoiding the loss of significant information that could result from pre-processing processes.
In section 4, we present the results obtained from our
study, demonstrating the efectiveness of the proposed
model in classifying environmental noise. The results show
a significant improvement in classification accuracy,
achieving remarkable performance in the tests performed. We will
also discuss the impact of a post-processing filter [
Environmental noise, defined as any unwanted sound
generated by the surrounding environment, is a major source
of noise pollution. These sounds may come from natural
sources such as sea waves or from man-made sources such
as vehicle trafic, alarms, voices and electronic devices.
Effective management of such noise requires methods that
go beyond simply measuring sound pressure levels (dB),
including characterising and identifying the type of noise [
Advances in Machine Learning (ML) and Deep Learning (DL) techniques have radically transformed the approach to data analysis, allowing us to discover unexpected complex patterns in audio data. In this study, we adopted an innovative methodology that exploits neural networks to directly process audio signals in .wav format. The aim is to evaluate the ability of these networks to accurately classify diferent sound events without resorting to pre-processing techniques that could compromise data integrity. In the subsection 3.1 we will describe the datasets used, the breakdown of these for training, validation and testing of the neural network and in 3.2 the CNN network used for the classification of ambient noise.
The dataset used in this research was composed by
merging three distinct public databases: UrbanSound [
Artificial intelligence (AI) represents a vast and evolving ifeld of study that aims to emulate human cognitive capabilities through the development of autonomous hardware and software systems. This ambition to reflect human intelligence in machines has led to the development of technologies capable of autonomous learning, adaptation, reasoning and planning. At the heart of AI are advanced algorithms and computational techniques, which make it possible to replicate typically human behaviours, such as interaction with the environment and decision-making. The applications of artificial intelligence range in diferent fields, from industrial to domestic, demonstrating its potential to improve both the activities of businesses and public administrations and the everyday lives of people.
Convolutional Neural Networks (CNNs) stand out for their efectiveness in analysing visual and sound data due to their ability to identify complex patterns through the use of convolutional filters.
Our CNN architecture follows a structured model starting with the input layer, proceeding through convolutional and activation (ReLU) layers, pooling, and culminating in a fully connected layer for final classification. This design allows the network to process audio features from the simplest to the most complex, facilitating deep and robust data learning. The detailed configuration of convolutional, pooling, and fully connected layers provides a powerful means to extract and interpret sound features, making CNNs particularly suitable for the recognition and classification of complex sound events. Our research aims to demonstrate the efectiveness of this approach in the field of acoustic analysis, contributing significantly to the field of signal processing and audio classification.
The neural network used in this study is based on the
architecture of 1D convolutional neural networks and, in
particular, on the “M5 (0.5M)" model described in [
The neural network’s input is a vector containing sequences of audio waveforms, each with a duration of = 2 seconds. The CNN neural network determines the index associated with one of = 14 diferent classes ( = 1, ..., ) using the LogSoftMax function. The network is trained by feeding RAW sequences representing diferent environmental noises.
The validation process of our approach was carried out through a rigorous experiment involving the direct input of raw audio data, in .wav format, into the convolutional neural network (CNN). Below, we present a detailed analysis of the performance obtained during the diferent phases of training, validation and testing of the network.
During the training phase, we observed a progressive improvement in network performance, as illustrated in Fig. 1. This graph shows an increase in accuracy and a decrease in the loss function as the epochs progress, highlighting the efectiveness of the learning process. The dataset was split into a proportion of 70% for training and 30% for validation, as illustrated in Section 3.1.
Fig. 2 presents the confusion matrix obtained from the validation of the model, providing a clear indication of its classification capability across the diferent sound categories.
The efectiveness of the model was further verified through testing on a separate dataset, achieving an impressive accuracy of 96.93%. Fig. 3 illustrates the confusion matrix for this phase, confirming the network’s high accuracy in recognising environmental sounds.
The introduction of a post-processing filter, called the
“recurrence filter" [
The results underline the efectiveness of our approach based on the use of convolutional neural networks for analysing environmental sound, highlighting the potential of deep learning techniques in overcoming the challenges of accurately recognising complex sound events.
This study introduced a new approach for the classification of environmental noise, exploiting the potential of Deep Learning techniques to address one of the most pressing challenges in signal processing and telecommunications. Through the use of a CNN trained on a unified dataset derived from three diferent public sources, it is shown that high accuracy in the classification of environmental noise events can be achieved without the need for complex preprocessing. The results obtained reveal a marked improvement in classification accuracy, highlighting the efectiveness of our model both in the testing phase and in the application of post-processing techniques. These results not only confirm the value of convolutional neural networks in acoustic analysis, but also open the way for future research to explore the applicability of such methods in broader areas, including urban noise monitoring and the improvement of telecommunication systems. In conclusion, our study contributes significantly to the body of research on signal processing, proposing an efective and eficient model for the classification of ambient noise, with direct implications for environmental sustainability and quality of life in urban areas.