Blind source separation by long-term monitoring: A variational autoencoder to validate the clustering analysis

Abstract

Noise exposure influences the comfort and well-being of people in several contexts, such as work or learning environments. For instance, in offices, different kind of noises can increase or drop the employees' productivity. Thus, the ability of separating sound sources in real contexts plays a key role in assessing sound environments. Long-term monitoring provide large amounts of data that can be analyzed through machine and deep learning algorithms. Based on previous works, an entire working day was recorded through a sound level meter. Both sound pressure levels and the digital audio recording were collected. Then, a dual clustering analysis was carried out to separate the two main sound sources experienced by workers: traffic and speech noises. The first method exploited the occurrences of sound pressure levels via Gaussian mixture model and K-means clustering. The second analysis performed a semi-supervised deep clustering analyzing the latent space of a variational autoencoder. Results show that both approaches were able to separate the sound sources. Spectral matching and the latent space of the variational autoencoder validated the assumptions underlying the proposed clustering methods.

Statistical analysis of sound level meter monitoring

“…Previous works confirmed a s.d. equal to 5 dB as a good threshold to separate and identify a mechanical noise from a human source, i.e., speech [4,7]. Thus, scenarios a and b depict the condition with two mechanical sources; scenarios c and d are constituted by the interactions of a mechanical source and a human source; scenarios e and f represent the case with two human sources.…”

“…Two unsupervised techniques, the Gaussian Mixture Model (GMM) and the K-means clustering (KM), has been used in offices and university lecture halls to identify, separate and measure different kinds of noise sources [4][5][6]. A deep learning approach, made through a variational autoencoder, assessed the GMM as the best algorithm to perform this kind of analyses [7].…”

“…Figure1shows two examples of SPLs occurrences collected during a long-term monitoring carried out in two different contexts: a office and a university lecture hall. According to previous studies, the GMM results to be the most flexible and reliable algorithm to analyze a sound level meter long-term monitoring[7]. Thus, it is possible to create synthetic acoustic scenarios combining different Gaussian curves.…”

A Novel Blind Audio Source Separation Utilizing Adaptive Swarm Intelligence and Combined Negentropy-Cross Correlation Optimization

“…It details the development of a unique cost function and the use of Newton gradient descent for optimizing the demixing matrix. The research work conducted by Salvio et al [33] discusses the impact of noise on workplace productivity and wellbeing, presenting a dual clustering approach to sound source separation using machine and deep learning techniques. The study utilizes long-term sound data, applying Gaussian mixture models and semi-supervised deep clustering to effectively differentiate between traffic and speech noises, highlighting the practical application and validation of these techniques in real-world settings.…”