Statistical effects of selected noise characteristics on speaker recognition in automotive environments
Sven Tuchscheerer, Christian Kraetzer, Jana Dittmann, Tobias Hoppe · 2011
A statistical analysis using the univariate, multifactorial analysis of variance (ANOVA) is used in this paper to investigate the impact of selected noise characteristics (here a 4-factorial design: amplitude, complexity, harmony and fundamental frequency) to speech signals and consecutively to the detection performance in speaker recognition systems (exemplarily used here: the BioSecure reference system ALIZE) in automotive application scenarios. An application scenario specific set of noise signals is recorded and generated and used to evaluate the influence of the noise characteristics. The results show that especially the amplitude and the fundamental frequency show a significant impact (p-values 0.5).