Title: A quantitative analysis of hands-free speech enhancement using real automobile data
Publisher: IEEE Global Conference on Signal and Information Processing (GlobalSIP)
Abstract: This paper provides a detailed comparison study between three different vehicles’ Bluetooth built-in noise cancellation filter with two widely used techniques in speech enhancement, Spectral Subtraction (SS) and Wiener filtering (WF). The main purpose is to determine if any of these two filters provide superior audio quality over the built-in filter. In literature, several authors have compared the performance of SS to that of WF using, primarily, simulated data, whereas this paper uses real-time data samples collected from cars subjected to a noisy environment with varying sound levels in search of an optimal solution. The cars were driven at different speeds with the windows and fan set to different configurations. In the process of comparing the three noise-canceling algorithms, the collected data were filtered using each filter, and the resulting tracks were analyzed both subjectively and objectively. Overall, SS outperformed WF by canceling more noise and/or conserving speech related frequency peaks. In all cases, but one, both Wiener and SS filters outperformed the built-in filter. The audio tests analyzed subjectively agree with the plot results.