Improving In-Vehicle Voice Recognition Systems: Basis for Enhancing the Telematics Experience

Although voice recognition is not essential for telematics application, it is certainly one of the most attractive features for the customer who buys it. The appealing is such that may determine the whole system success in the market and, therefore, deserves our attention. This paper discusses the voice recognition task and ways to improve it in the vehicle hostile environment. In order to create a robust system, artificial neural networks are being used in combination to the appropriate signal processing, which includes temporal alignment and spectral transformation for MFCCs features extraction. The problems related to in-vehicle application are described and feasible solutions are suggested. A proposed architecture is then designed considering all foreseen complications, taking advantage of sensors already available in other vehicle systems, ensuring the performance optimization at minimum cost increase.