The rapid advancement of two-wheeler automotive technology has underscored the need for enhanced connectivity and user interaction systems. This thesis explores the integration of Automatic Speech Recognition (ASR) systems into an Android Automotive-Based Embedded Connectivity Platform specifically designed for TwoWheelers, aiming to improve safety and user experience through hands-free operation. The study focuses on selecting and implementing suitable ASR models and computational strategies that operate efficiently within the constraints of an embedded system. The research evaluates different computational approaches, namely edge computing, cloud computing, and a hybrid strategy, to determine the most effective solution based on metrics such as transcription accuracy, processing time, and resource utilization. Two distinct ASR integration strategies were developed and tested: an edge computing based application and a cloud computing based application. Performance evaluations were conducted using a comprehensive test database, with a focus on transcription time relative to input audio duration, CPU usage, and transcription accuracy under varying noise conditions. Additionally, the impact of Signal-to-Noise Ratio on transcription accuracy was analyzed, employing advanced metrics to assess the robustness of the ASR systems. The findings indicate that cloud computing outperforms edge computing in both accuracy and processing efficiency, making it the recommended strategy for this application. However, the thesis also proposes a hybrid approach, combining edge and cloud computing, to optimize performance and scalability. This research contributes to the field of automotive connectivity by providing a detailed analysis of ASR integration strategies and recommending a hybrid strategy to enhance the performance and scalability of ASR systems in real-world applications.
Study of Computational Strategies to Integrate Automatic Speech Recognition in an Android Automotive-Based Embedded Connectivity Platform for Two-Wheelers
Soto Argüello, M. A. (Author). 2024
Student thesis: Master's Thesis