| KEYWORD |
Area Architecture
Blind estimation of acoustical parameters from signals recorded in the room
keywords ACOUSTIC RESPONSE, ACOUSTICS, ARTIFICIAL INTELLIGENCE, DEEP LEARNING, NEURAL NETWORKS
Reference persons ARIANNA ASTOLFI, ELIANA PASTOR, ANTONIO SERVETTI, LOUENA SHTREPI
Research Groups TEBE
Thesis type MACHINE LEARNING
Description The acoustical quality of a room for speech or music is investigated with the measurements of acoustical parameters obtained from the room impulse response in unoccupied conditions. This requires an omnidirectional source and a microphone moved in different positions. The procedure is heavy in terms of equipment and time spent on the measurements and analysis. Furthermore, it must be carried out in an unoccupied room, which does not represent a realistic listening condition. The thesis will explore the methodology of blind estimation of room acoustical parameters from speech or music signals directly recorded in the room (i.e., offices, classrooms, restaurants, concert halls, etc.), thus reducing the time spent for the measurements and the cost of the equipment. It will be explored the possibility of estimating the Speech Transmission Index (STI) and other parameters recommended in ISO 3382-1 standard, with high accuracy. Deep neural networks, trained from a massive number of acoustic signals will be used to estimate the acoustical parameters for seven-octave bands. The Thesis will be carried out in collaboration between the Applied Acoustics Lab of the Energy Department (DENERG) and the Department of Control and Computer Engineering (DAUIN).
See also 1-s2.0-s0003682x21004667-main.pdf
Required skills MACHINE LEARNING FUNDAMENTALS
Deadline 03/05/2025
PROPONI LA TUA CANDIDATURA