Excerpt from Methods section of: “A Multi-loudspeaker Binaural Room Impulse Response Dataset with High-Resolution Translational and Rotational Head Coordinates in a Listening Room” - arXiv:2403.12258 


2.1 Data Collection We measured BRIRs in an irregular-shaped listening room with a near-shoebox shape (see Fig. 1 for the exact dimensions). The listening room has a RT60 of 0.24 s averaged in the range between 1300 and 6300 Hz. The room floor is covered with carpet, and the walls and ceiling are partially covered with acoustic panels. Fig. 1 also shows the setup and dimensions of the measurement system. A linear array of eight loudspeakers was used as sound sources, and each is a Focal Shape 40 4-inch Flax woofer. The loudspeaker array layout was initially intended for sound field control applications, such as rendering personal sound zones. A Br¨uel & Kjær Head and Torso Simulator (HATS, Type 4100) was used as the mannequin listener, with its built-in microphones replaced with a pair of in-ear binaural microphones (Theoretica Applied Physics BACCH-BM Pro). The microphones were calibrated and free-field equalized before the measurement. A custom-made, computer-controlled mechanical translation platform was applied to enable translational movements, and a turntable (Outline ET250-3D) was mounted on top of the platform for rotational movements in the azimuth. The BRIR measurement grid has a range of [0.5, 1.0] m in the y direction (front/back) and [-0.5, 0.5] m in the x direction (left/right), with a 0.05 m spacing between adjacent grid points. The distances are relative to the center of the loudspeaker array. At each grid point, the BRIRs were measured at 37 different azimuth angles from the listener facing left to facing right, with a 5◦ spacing between adjacent angles. In total, there are 68376 (= 11 y-translations × 21 x-translations × 37 azimuthal rotations × 8 loudspeakers) BRIRs measured. We measured BRIRs by playing back exponential sine sweep (ESS) signals from the loudspeakers and recording the signals received with binaural microphones. Each sine sweep signal has a length of 500 ms at a 48 kHz sampling rate and is generated using the synchronized ESS method [20], with a start frequency of 100 Hz and an end frequency of 24 kHz. The synchronized ESS method is a variant of the traditional ESS method [21], with the advantage of correctly estimating higher harmonic frequency responses. All eight loudspeakers were triggered in series with no overlapping between the ESS signals. The entire data collection process was split into multiple measurement sessions. For each session, we manually fixed the distance from the listener to the array in the y direction and automated the movements in the x direction and the azimuthal rotations. The measurement automation, signal generation, and data collection were implemented in Cycling ’74 Max 8. The BRIR post-processing was performed in MATLAB. Each session lasted for approximately 2 hours, and the entire data collection process took 9 days.

2.2 Data Processing The BRIRs were obtained by first deconvolving the recorded signals with the ESS signal in the frequency domain, with 32768-length FFT at a 48 kHz sampling rate. Then, a fourth-order highpass Butterworth filter with a cutoff frequency of 100 Hz was applied to the deconvolved signals to remove the low-frequency noise that was present during the measurement. Finally, the deconvolved signals were truncated to the first 16384 samples (corresponding to 341.3 ms) and globally normalized. No loudspeaker equalization was applied to the BRIRs as the loudspeaker-specific information, such as directivity, is an integral part of the BRIR and therefore is difficult to compensate for. The processed BRIRs, together with the corresponding listener position and orientation coordinates, were saved as separate files corresponding to different y translations in the SOFA (spatially-oriented format for acoustics, [22]) format, following the AES69-2022 (SOFA 2.1) Standard. The dataset was generated using SOFA Toolbox for MATLAB/Octave version 2.2.0.