Research and Develop Solutions to Traffic Data Collection Based on Voice Techniques

This paper addresses two primary challenges within the context of the current intelligent traffic system, Urban Traffic Estimation System (UTraffic). Firstly, it endeavors to identify and explore an additional traffic data source, supplementing the two existing data sources utilized for training the Automatic Speech Recognition (ASR) model. Secondly, it aims to conduct experiments using the newfound dataset in conjunction with advanced ASR models to ascertain the most optimal ASR model for integration into UTraffic. The key methodologies employed to tackle these issues include collecting traffic reports from a radio station, processing the data for training ASR models, and experimenting with different ASR models. In essence, this research endeavor strives to generate an enhanced dataset comprising authentic real-world data, leading to superior ASR model accuracy compared to the presently deployed ASR model within UTraffic.