Article
Speech Recognition Predictions for Measured and Simulated Binaural Room Impulse Responses (en)
* Presenting author
Abstract:
There is a growing interest in realistic scenarios in hearing research, particularly for evaluating speech recognition. Complex auditory environments enable more ecologically relevant investigations than traditional methods. This contribution investigates the impact of different room acoustic models on predicted speech recognition thresholds (SRTs) in a living room. A real living room was virtually reproduced using the room acoustics simulator RAZR and the Toolbox for Acoustic Scene Creation and Rendering (TASCAR), varying the source directivity, the listener’s head direction, and the head model. Recorded and simulated binaural room impulse responses (BRIRs) were convolved with sentences of the Oldenburg sentence test for different spatial configurations of speech and noise sources, reflecting real-life communication scenarios. SRTs were predicted using the Binaural Speech Intelligibility Model.SRTs predicted with both room models show a high correlation (R²=0.95) and low bias (0.1 dB). The SRTs based on recorded and simulated BRIRs show a bias of 0.0-0.3 dB and correlations of R²=0.55-0.77, depending on the room model. Source directivity and the head model affect the SRTs by up to 3 and 2 dB, respectively. The study quantifies the impact of different simulation parameters on SRTs, which is important for interpreting speech recognition in acoustically complex scenarios.