Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Assessment of Deep transfer learning for Speech Transmission Index prediction
Submission Author:
Eriberto Oliveira do Nascimento , PR
Co-Authors:
Eriberto Oliveira do Nascimento, PAULO HENRIQUE TROMBETTA ZANNIN
Presenter: Eriberto Oliveira do Nascimento
doi://10.26678/ABCM.COBEM2021.COB2021-1087
Abstract
The presence of noise interferes negatively with the teaching practice. Thus, good acoustic conditions become a relevant issue in classrooms. This study aims to determine the significance of the following classroom conditions: background noise - (A), sound absorption coefficient - (B), confinement - (C), and occupation - (D) on the Reverberation Time (RT) and Speech Transmission Index (STI). Then, based on measurements in 5 classrooms and its validated simulations in the ODEON 11 software, a response matrix was created, totaling 80 virtual rooms using the Design of Experiments. The quantification of the input variable significance was determined using deep neural networks. The results showed that the higher the RT lower was the STI. The following composition explained the percentual variation of the RT: B, D, and C, while for the STI, the conditions were A, B, D. Therefore, in conclusion, the results obtained agree with the current literature.
Keywords
Deep learning, deep transfer learning, room acoustics, Speech Transmission Index, reverberation time
