The Influence of In-Vehicle Noise on Speech Recognition for Automotive Voice-Activated Control Systems

Voice activation provides a safe and convenient method of controlling vehicle systems such as in-car entertainment, telecommunications, and climate control. In the fullness of time it is likely that there will be a high demand across all vehicle classes for such systems for a wide range of vehicle control functions. One of the challenges facing their development and introduction into vehicles is that of the in-vehicle noise environment. This paper investigates the influence of in-vehicle noise on speech recognition performance for an automotive voice-activated system over a range of vehicle classes and typical vehicle operating conditions. The investigation was carried out in a controlled laboratory environment using pre-recorded speech and in-vehicle noise recordings. Analysis of the experimental data showed significant system performance sensitivity and well defined, consistent, performance thresholds to the overall level of noise and spectral balance components typically found in the automotive environment. Significant differences in speech recognition performance between the male and female voice were also found. With this information it has been possible to identify both the vehicle classes and operating conditions that are likely to create the most challenging environments for acceptable speech recognition performance. This provides a useful reference for further research and development of both voice-activated speech recognition (VSR) system software and hardware design, and identification of in-vehicle noise targets for future vehicles.