Saved in:
Bibliographic Details
Main Authors: Muhammad, Imran, Schuller, Gerald
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.24769
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915522069659648
author Muhammad, Imran
Schuller, Gerald
author_facet Muhammad, Imran
Schuller, Gerald
contents Accurate prediction of energy decay curves (EDCs) enables robust analysis of room acoustics and reliable estimation of key parameters. We present a deep learning framework that predicts EDCs directly from room geometry and surface absorption. A dataset of 6000 shoebox rooms with realistic dimensions, source-receiver placements, and frequency-dependent wall absorptions was synthesized. For each configuration we simulate room impulse responses (RIRs) using Pyroomacoustics and compute target EDCs. Normalized room features are provided to a long short-term memory (LSTM) network that maps configuration to EDC. Performance is evaluated with mean absolute error (MAE) and root mean square error (RMSE) over time. We further derive early decay time (EDT), reverberation time (T20), and clarity index (C50) from predicted and target EDCs; close agreement is observed (e.g., EDT MAE 0.017 s, T20 MAE 0.021 s). The approach generalizes across diverse rooms and supports efficient room-acoustics modeling for early-stage design and real-time applications.
format Preprint
id arxiv_https___arxiv_org_abs_2509_24769
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Deep Learning-Based Prediction of Energy Decay Curves from Room Geometry and Material Properties
Muhammad, Imran
Schuller, Gerald
Audio and Speech Processing
Accurate prediction of energy decay curves (EDCs) enables robust analysis of room acoustics and reliable estimation of key parameters. We present a deep learning framework that predicts EDCs directly from room geometry and surface absorption. A dataset of 6000 shoebox rooms with realistic dimensions, source-receiver placements, and frequency-dependent wall absorptions was synthesized. For each configuration we simulate room impulse responses (RIRs) using Pyroomacoustics and compute target EDCs. Normalized room features are provided to a long short-term memory (LSTM) network that maps configuration to EDC. Performance is evaluated with mean absolute error (MAE) and root mean square error (RMSE) over time. We further derive early decay time (EDT), reverberation time (T20), and clarity index (C50) from predicted and target EDCs; close agreement is observed (e.g., EDT MAE 0.017 s, T20 MAE 0.021 s). The approach generalizes across diverse rooms and supports efficient room-acoustics modeling for early-stage design and real-time applications.
title Deep Learning-Based Prediction of Energy Decay Curves from Room Geometry and Material Properties
topic Audio and Speech Processing
url https://arxiv.org/abs/2509.24769