Speech Recognition for On-Site Collection of License Plate Data: Exploratory Application Development and Testing

Abstract

The collection of traffic data in the field, both from manual and automated methods, is vital to the management and maintenance of the transportation infrastructure. Historically, many methods for manual data collection have been utilized, ranging from pencil and paper to portable computers. At present, with advances in computing performance and technology over the last several years, speech recognition technology has emerged as a potentially viable tool for data collection, a tool with very distinct advantages over some existing methods of manual data collection. This paper discusses the development and testing of a speech recognition enabled software program for data collection in the field. The specific program developed was designed to enable the collection of vehicular license plate data from roadside observation locations. This particular application proved to be a very demanding test for speech recognition due to several factors—loud background noise conditions at roadside locations, the need for high accuracy rates, and the need for sufficient computational speed to handle potential high sampling rates. Many field tests were conducted and key measures of effectiveness were compared between the speech entry method and a method utilizing portable computer keyboard-based data entry. Overall, the results showed promise that speech recognition could become a method that will improve the ease and efficiency of collecting transportation data in the field, which should also lead to increased data quantity and quality.