Binaural room impulse responses for speech-in-noise testing in virtual restaurants

Binaural room impulse responses (BRIRs) were recorded in two real restaurants using the tone-sweep method (Farina, 2007). In a semi-automated procedure, ten-second logarithmic tone sweeps were played from Cambridge Minx satellite speakers and recorded binaurally using a B&K 4100 head and torso simulator. These were convolved with an inverse sweep to derive BRIRs between different seating positions. In each restaurant, a speaker position and an opposing listener position were selected for each table. BRIRs were recorded from each table to every other. In Mezza Luna, with 18 tables, three listener head positions (-30°, 0°, +30° with respect to the speaker position on the same table) were also recorded, making 18×18×3=972 BRIRs. In A Caverna, with 15 tables, 15×15=225 BRIRs were recorded. Predicted speech intelligibility was explored using the Jelfs et al. (2011) model of spatial release from masking for speech. Predicted SRTs (pSRTs) improved when the head was turned to ±30°. In Mezza Luna, a long narrow restaurant, pSRTs were higher in the middle of the room. In A Ceverna, a restaurant on two levels, pSRTs were much lower on the mezzanine floor. This set of predictions provides a powerful test for the model that can be evaluated using virtual acoustic listening tests based on the same BRIRs. The BRIRs could also be used to assess the acoustic capacity of the restaurants, the largest number of simultaneous voices in the room that listeners can tolerate.

Farina, A. (2007). Advancements in impulse response measurements by sine sweeps. Proceedings of 122nd AES Convention, Vienna.
Jelfs, S., Lavandier, M. and Culling, J. F. (2011). Revision and validation of a binaural model for speech intelligibility in noise. Hear Res, 275, 96-104.