BRUDEX Database: Binaural Room Impulse Responses with Uniformly Distributed External Microphones

There is an emerging need for comparable data for multi-microphone processing, particularly in acoustic sensor networks. However, commonly available databases are often limited in the spatial diversity of the microphones or only allow for particular signal processing tasks. In this paper, we present a database of acoustic impulse responses and recordings for a binaural hearing aid setup, 36 spatially distributed microphones spanning a uniform grid of (5x5) m^2 and 12 source positions. This database can be used for a variety of signal processing tasks, such as (multi-microphone) noise reduction, source localization, and dereverberation, as the measurements were performed using the same setup for three different reverberation conditions (T_60≈{310, 510, 1300} ms). The usability of the database is demonstrated for a noise reduction task using a minimum variance distortionless response beamformer based on relative transfer functions, exploiting the availability of spatially distributed microphones. An example how to load a impulse responses corresponding to the 'low' reverberation condition for the speaker located at 60 deg using MATLAB: dataStruct = loadRIR('low',60,1,<basePATH>);%<basePATH>: path where database is located on local machine For further MATLAB examples, please consider "wrapper_loadDataFromDB.m" in the "matlabScripts.zip" archive file. An example how to load a impulse responses corresponding to the 'low' reverberation condition for the speaker located at 60 deg using Python: dataloader = wrapper.BRUDEXDataloader()# <basePATH> is implicitly set to that path, where the file "wrapper.py" is located on local machine dataStruct = dataloader.load_rir(reverberation_condition='low', direction_of_arrival = 60, ha_av = 1,e_mic_run = None) For further Python examples, please consider "main.py" in the "pythonScripts.zip" archive file. Caution: We noticed some problems with the download of the databse when using the command line (e.g., via the zenodo_get, wget, or curl commands). These problems don't seem to appear when downloading the files with the "Download" buttons on the website instead. Reference: D. Fejgin, W. Middelberg, and S. Doclo, “BRUDEX database: Binaural room impulse responses with uniformly distributed external microphones,” in Proc. ITG Conference on Speech Communication, Aachen, Germany, Sep. 2023, pp. 1–5. @InProceedings{Fejgin2023, author = {D. {Fejgin} and W. {Middelberg} and S. {Doclo}}, booktitle = {Proc. ITG Conference on Speech Communication}, title = {{BRUDEX} Database: Binaural Room Impulse Responses with Uniformly Distributed External Microphones}, pages = {1-5}, month = {Sep.}, year = {2023}, address = {Aachen, Germany} }

当前，多麦克风处理领域对可比对数据集的需求日益凸显，尤其在声学传感器网络场景中尤为突出。然而，现有公开数据库往往存在麦克风空间多样性不足的问题，或仅支持特定类型的信号处理任务。

本文提出了一款面向双耳助听器配置的声学冲激响应与录音数据集：数据集包含36个空间分布式麦克风，部署于5m×5m的均匀网格中，同时设置了12个声源位置。由于本次测量针对三种不同混响条件（T₆₀≈{310、510、1300} ms）采用了完全一致的实验配置，该数据集可被用于多种信号处理任务，例如（多麦克风）降噪、声源定位以及去混响等。

本文基于相对传递函数的最小方差无失真响应波束形成器，针对降噪任务验证了该数据集的可用性，充分利用了空间分布式麦克风的部署特性。

以下为使用MATLAB加载声源位于60°方向、低混响条件下冲激响应的示例代码：
matlab
dataStruct = loadRIR('low',60,1,<basePATH>);
% <basePATH>：数据集在本地机器上的存储路径

如需更多MATLAB示例，请查阅压缩包`matlabScripts.zip`中的`wrapper_loadDataFromDB.m`文件。

以下为使用Python加载声源位于60°方向、低混响条件下冲激响应的示例代码：
python
dataloader = wrapper.BRUDEXDataloader()
# <basePATH> 默认为本地机器上`wrapper.py`文件所在路径
dataStruct = dataloader.load_rir(reverberation_condition='low', direction_of_arrival = 60, ha_av = 1, e_mic_run = None)

如需更多Python示例，请查阅压缩包`pythonScripts.zip`中的`main.py`文件。

注意：我们发现通过命令行（如使用zenodo_get、wget或curl命令）下载该数据集时存在部分问题，而通过网站上的「下载」按钮下载文件则不会出现此类问题。

参考文献：
D. Fejgin, W. Middelberg, 和 S. Doclo, "BRUDEX数据集（BRUDEX Database）：带有均匀分布外置麦克风的双耳房间冲激响应数据集", 载于《ITG语音通信会议论文集》, 德国亚琛, 2023年9月, 第1–5页。
bibtex
@InProceedings{Fejgin2023,
author = {D. {Fejgin} and W. {Middelberg} and S. {Doclo}},
booktitle = {Proc. ITG Conference on Speech Communication},
title = {{BRUDEX} Database: Binaural Room Impulse Responses with Uniformly Distributed External Microphones},
pages = {1-5},
month = {Sep.},
year = {2023},
address = {Aachen, Germany}
}

搜集汇总

背景与挑战

背景概述

BRUDEX数据库是一个专注于双耳房间脉冲响应的数据集，包含36个均匀分布在5x5平方米网格上的外部麦克风和12个声源位置，测量了三种不同混响条件（T_60≈310、510、1300毫秒）。该数据集旨在支持多麦克风信号处理任务，如噪声降低、源定位和去混响，适用于声学传感器网络和助听器研究，数据量约为200 GB，并提供MATLAB和Python的加载示例。