MIRaGe is a database that has been created as a joint project of the ASAP group and of prof. Sharon Gannot and people of the acoustic lab at Bar Ilan University (BIU), Israel. The database consists of multi-channel recordings performed in the BIU acoustic lab with adjustable reverberation time. The recordings provide information about room impulse responses (RIR) for various positions of a loudspeaker. In particular, the main positions correspond to 4104 vertices of a cube-shaped dense grid within a 46×36×32cm volume. The database thus provides a tool for detailed analyses of beampatterns of spatial processing methods as well as for training and testing of mathematical models of the acoustic field and of the room impulse responses.
Tensors – multidimensional arrays of impulse responses created by simulator and from MIRaGe
Audio-visual application from WASPAA 2019 – MVDR/MPDR beampatterns comparing RIR simulator and MIRaGe
J. Čmejla, T. Kounovský, S. Gannot, Z. Koldovský, P. Tandeitnik, “MIRaGe: Multichannel Database Of Room Impulse Responses Measured On High-Resolution Cube-Shaped Grid In Multiple Acoustic Conditions”, arXiv:1907.12421 [eess.AS], July 2019.
Download links contain zip files with the database itself and software for easier manipulation with the database.
Database zip files are divided into three categories (“WhiteNoise”, “Chirp” and “Misc”). The Whitenoise and Chirp categories contain recordings from the grid and OOG positions for the respective type of the excitation signal. Each zip file contains all recordings for a single microphone array in one reverberation level. For example “Chirp_100_01.zip” contains recordings of
the chirp excitation signal for microphones 1 through 5 (array 1) in 100ms of room reverberation time. The “*_on_SPK_mic.zip” files contain recordings for microphone 31 (close to the loudspeaker). User can select and download only those zip files containing data needed for the specific experiment (one array ~12 GB).
Unpacking the database:
Each zip file has the same folder structure. After downloading one or more zip files, unpack all of their contents into the same root folder.
Using the database (software):
The user can manually obtain all desired ATFs and RTFs through a single function called “getRTF”. See MATLAB function documentation (“help getRTF”) for more info. Alternatively, we provide a simple MATLAB GUI (“MIRaGe_GUI.m”) for easy automated mass RTF calculation.
The recording setup is situated in the acoustic laboratory, which is a 6x6x2.4 m rectangular room. A loudspeaker emulating the source is located within a cube-shaped volume of dimensions 46x36x32 cm (referred to as GRID) as shown the following figures.
The positions of the loudspeaker form a grid sampled every 2 cm across the x-axis and y-axis and every 4 cm across the z-axis, so there are 24x19x9=4104 possible source positions (grid vertices) in total. We use the Matlab-like notation to denote positions; for example, [:,:,1] means all vertices in the first horizontal slice of the grid.
In addition, there are 25 source positions located outside of the grid (OOG); 9 positions are close to the grid and 16 positions are situated along the walls. The center of the grid (the 5th level), as well as the OOG positions, were positioned at the same height of 115 cm. In the grid positions, the loudspeaker is directed in parallel to the y-axis towards the opposite corner of the room. In the OOG positions, the loudspeaker is directed towards the center of the room.
The entire setup is recorded by six static linear microphone arrays and one microphone mounted 2 cm in front of the loudspeaker, which is changing its position with the loudspeaker (microphone 31). Three microphone arrays are placed directly in front of the grid at the distance of 1, 2, and 3 m from the center of the grid. The other three arrays are located at the angle of -45 degrees at the same distances. All arrays are directed towards the grid centre and placed at the same height of 115 cm. Each array consists of 5 microphones with the inter-microphone spacing of -13, -5, 0, +5 and +13 cm relative to the central microphone.
The sidewalls and the ceiling of the acoustic lab consist of revolving double-sided panels, by which the reverberation time of the room can be controlled. We have chosen three reverberation time levels: 100, 300, and 600 ms.
The positioning of the loudspeaker within the GRID was realized using a precise three-axis positioning system. It consists of a 2D plotter and a lift table.
For each position, two excitation signals were played and recorded in sequence: The first signal (Chirp) consists of two repetitions of a logarithmic swept-frequency cosine signal with a total length of 20 seconds (0.5 s silence, 8 s chirp, 2 s silence, 8 s chirp, 1.5 s silence). White noise (WN) was used as the second excitation signal with a total length of 10 seconds (0.5 s silence, 8 s WN, 1.5 s silence).
We have also recorded one hour of room tone (silence) and one hour of diffuse babble noise for each reverberation time setting. The babble noise was simulated by using eight loudspeakers each playing a different multi-speech sequence and each placed approximately 1 m from the walls: one in each corner and one in the middle of each wall. The loudspeakers were directed towards the nearest corner or the nearest wall.