The design and development of a system for controlling automotive functions using speech recognition

Abstract

As car manufacturer installed more electronic control interfaces like Wifi, Bluetooth or infrared systems, the maneuverability and accessibility of the automobile itself is enhanced. We opted for personal digital assistant (PDA) which has the capabilities of a computer yet compact enough to be mounted on car’s dashboard. The PDA is equipped with automatic speech recognition (ASR) system, thus providing the mean for activating devices via voice. Other benefits gained from using PDA includes easy upgradeability of the ASR engine and resolving the portability issue as consumers may own more than a single car. Apart from the embedded Artificial Neural Network (ANN) based ASR engine, we developed a prototype engine denoted as Support Vector Machine-Dynamic Shifting Window (SVM-DSW) to accommodate speaker independent mode. SVM-DSW is targeted for applications that demand accuracy and reliability. Accuracy gets top priority in high risk tasks such as driving or piloting, surgical procedures, etc. where the slightest error gives disastrous consequences. Consistency is also of paramount prerequisite because the accuracy has to be reproducible time and time again without failure. SVM-DSW has both of these qualities as well as being low in computational cost (using whole word recognition unit and embedded grammar rule) allows it to be ported into Very Large Scale Integration (VLSI) technology. Voice activated household appliances could also benefit from such integration.