Analysis of vehicle voice recognition performance
Analysis of vehicle voice recognition performance in response to background noise and gender-based frequency.
To address customer complaints on voice recognition (VR), Hyundai and Brüel & Kjær Global Engineering Services analysed the performance of the in-vehicle VR system and the sensitivity to background noise and gender-based frequency.
Connected and smart devices are all around us in our daily life and their use is anticipated to increase exponentially over the years. Most of these devices are controlled by voice-activated assistants, and the quality of the VR system can be affected by multiple parameters.
“With speech recognition becoming the go-to technology for the automotive industry customers, the urgency to improve the recognition rate was so important. This project addressed that need, and the outcome was very beneficial to Hyundai and added new insight in terms of the recognition performance and human behaviour.”Rasheed Khan, Manager, Multimedia Validation & Product Quality, Hyundai-Kia America Technical Center, Inc.
For vehicles, VR is constantly reported as a major quality issue. For the 2017 model year, market research indicates that it is again the most frequent problem reported in vehicle dependability studies. In a vehicle, the performance of VR is affected by multiple parameters, including the background noise, the placement of the microphone and the voice command itself. In addition, the gender of the speaker plays a role too, as it differentiates the speech pattern and the frequency content.
The North American team of Hyundai performed an R&D project together with Brüel & Kjær Global Engineering Services to analyse the performance of in-vehicle VR. The team defined a process based on actual vehicle tests to identify the sensitivity of the VR performance with respect to background noise and speaker gender. The results showed, as one would expect, that background noise can be detrimental to VR, but that the effect can be reduced with proper tuning of the noise cancellation algorithm. The team also found out that for specific vehicles, there is a significant reduction in the recognition rate for female speakers compared to male speakers. For these investigations, the team developed a consistent process for lab-based VR evaluations that can be used to assist with the tuning and calibration of VR systems.