A binaural room impulse response (BRIR) describes the characteristics of acoustic wave interactions from a sound source in a room to the torso, head, and ears of a listener. The use of BRIRs has been ubiquitous in many audio applications. For example, in spatial audio reproduction with headphones, BRIRs are used as audio filters to simulate or reproduce an immersive and perceptually plausible sounding environment; in loudspeaker-based applications, the frequency-domain counterparts of BRIRs are equivalent to the acoustic transfer functions between the loudspeakers and the listener’s ears, based on which audio filters are designed for tasks such as crosstalk cancellation (Cooper and Bauck, 1989; Gardner, 1998; Choueiri, 2018), room correction/loudspeaker equalization (Karjalainen et al., 1999; Lindfors et al., 2022), and personal sound zones (Druyvesteyn and Garas, 1997; Betlehem et al., 2015; Qiao and Choueiri, 2023a). In addition to audio reproduction and rendering, BRIRs have also played an important role in other audio-related tasks, such as sound source localization (Shinn-Cunningham et al., 2005), sound source separation (Yu et al., 2016), and audio-visual learning (Younes et al., 2023).

As implied by its name, a BRIR is dependent on both the listener’s anthropometric features (e.g., ear size and shape) and the room’s geometry and acoustic properties. Due to the complex acoustic interactions, such as sound reflections in the room and scattering off the listener, a BRIR varies with both the listener’s position and orientation in the room. This is unlike room impulse response (RIR), which only depends on the position, or anechoic head-related impulse response (HRIR) which, in the far-field case, only depends on the orientation. Although there have been multiple HRIR and RIR datasets (Sridhar et al., 2017; Brinkmann et al., 2019; Koyama et al., 2021) available, and while it is possible to synthesize BRIRs from HRIRs and RIRs using methods such as the image source model (Wendt et al., 2014), synthesized BRIRs lose physical accuracy and can only maintain perceptual plausibility. While this is sufficient for some applications, such as headphone-based auralization, it is not appropriate for others, such as crosstalk cancellation and personal sound zones, where measured BRIRs are required. Moreover, the lack of a high-resolution BRIR dataset, compared to existing ones measured at sparse listener positions and orientations (Jeub et al., 2009; Kayser et al., 2009; Erbes et al., 2015), limits the study of BRIR modeling and interpolation at high frequencies and the development of machine learning-based audio applications.

In this paper, we introduce a dataset that contains BRIRs measured in an acoustically treated listening room using multiple loudspeakers and at high-resolution translational and rotational head coordinates. Although the dataset is only measured in one specific room, it is expected that it will be useful for a wide range of applications as it faithfully captures the spatial dependency of BRIRs on listener positions and orientations. For example, it can be directly applied to studies of BRIR modeling and to interpolation and applications that require multi-loudspeaker BRIRs, such as crosstalk cancellation and personal sound zones. It has been shown (Qiao and Choueiri, 2023b) that the spatial sampling resolution of the dataset is adequate for rendering personal sound zones with continuous listener movements within a certain frequency range. With proper interpolation between the measured BRIRs, the dataset can also be used to simulate binaural audio for headphone-based auralization with continuous listener movements. In addition, the dataset can be used for either data augmentation or performance evaluation in a wide range of machine learning-based tasks that require binaural audio with listener movements in multiple degrees of freedom.

We examine the dataset by visualizing 1) the time index of the BRIR onset, 2) the peak amplitude of the BRIR, and 3) the interaural time difference (ITD) of the BRIR as functions of the listener position. Both the onset and the ITD of the BRIRs were calculated in a thresholding approach (see Katz and Noisternig (2014) as an example). Figure 2 shows the onset, peak amplitude, and ITD of the BRIRs measured at the fourth loudspeaker (counting from the left) and with the listener facing forward. The colors in the figures were interpolated using the “interp” option of the MATLAB “pcolor” function. All three figures show clear spatial dependency on the listener’s position, with the onset increasing and the peak amplitude decreasing as the listener moved away from the loudspeaker. The ITD was nearly zero when the listener was on-axis with the loudspeaker and increased as the listener moved to off-axis positions. Note that the maxima of the peak amplitude in Figure 2 are not on-axis with the loudspeaker, which was due to the occlusion effect of the listener’s head and ear.