Multi-room Apartments Simulation (MRAS) Dataset

Overview

The Multi-Room Apartments Simulation (MRAS) dataset is a novel multi-modal dataset created specifically for the task of estimating spatially-distributed acoustic parameters in complex scenes. It includes a large collection of scene geometries and Room-impulse Responses (RIRs), from dozens of unique source positions and a dense grid of receivers. The scene geometries are generated algorithmically by connecting shoe-box rooms using two distinct patterns. As a part of our publication, “Scene-wide Acoustic Parameter Estimation” (see citation below), we release the MRAS dataset for public use. This site includes the 3D meshes and the raw simulated RIRs; the pre-processed floormaps, acoustic parameters, and acoustic maps used in our experiments, along with accompanying code, can be found at https://github.com/facebookresearch/SceneAcousticEstimation.

Details

Scene Generation

A total of 1,000 distinct acoustic scenes are generated by creating 100 unique geometries with both linear and grid patterns, each assigned five different sets of materials. The linear pattern connects shoe-box rooms along shared boundaries, while the grid pattern subdivides a large area into multiple connected rooms, with varying room heights and randomly assigned materials sampled from realistic options, plus highly absorptive and highly reflective types. Doorframes between rooms have random widths, allowing for wide hallways, and the algorithmic construction results in geometries with high acoustical complexity.

Acoustic Simulation

The dataset has approximately 4 million RIRs, divided across the 1000 scenes. For each scene, 3 receiver positions per room are uniformly sampled to act as source positions, regardless of the room size. We use a dense grid of receivers with 0.3 m of spacing, at least 0.5 m away from any boundary. The RIRs are provided in 2nd-order ambisonics format.

License

CC BY 4.0

Citation

If you use this dataset, or the accompanying processed data, models, or code, please cite:

@inproceedings{falconperez2025,
 author = "{Falcón Pérez}, Ricardo and Gao, Ruohan and Mueckl, Gregor, and 
{Amengual Gari}, Sebastia V. and Ananthabhotla, Ishwarya",
 year = {2025},
 title = {Scene Wide Acoustic Parameter Estimation},
 booktitle = {IEEE Workshop on Applications of Signal Processing to Audio and 
Acoustics (WASPAA)},
 }