GIR dataset: A geometry and real impulse response dataset for machine learning research in acoustics

Autor: Achilleas Xydis, Nathanaël Perraudin, Romana Rust, Kurt Heutschi, Gonzalo Casas, Oksana Riba Grognuz, Kurt Eggenschwiler, Matthias Kohler, Fernando Perez-Cruz
Jazyk: angličtina
Rok vydání: 2023
Předmět:
Zdroj: Applied Acoustics, 208
ISSN: 0003-682X
Popis: Acoustics play a significant role in our everyday lives, influencing our communication, well-being, and perception of space. Fast and precise acoustics simulation is crucial for the accurate design of real spaces by architects and acousticians and maximises the user's immersion in virtual and augmented reality environments. Computer simulation techniques can help to simulate and analyse acoustics. However, their cumbersome, computationally expensive, and often inaccurate results discourage most architecture practices from including acoustic evaluation in their design workflow and prevent real-time accurate audio synthesis in virtual reality. Recent advancements in Machine Learning (ML) and particularly Deep Learning offer compelling solutions to address the above problems. ML methods require large datasets for training, and existing datasets are either not large enough, contain synthetic data, or are not suitable for room acoustics research. This paper presents the GIR Dataset, a dataset of 920’712 real Impulse Responses (IRs) of 312 architectural geometries for the study of early reflections from diffusive surfaces. The paper provides a detailed description of the GIR Dataset's content and an ML use-case example. The dataset and the code described in this paper are open-sourced.
Applied Acoustics, 208
ISSN:0003-682X
Databáze: OpenAIRE
Externí odkaz: