Road Construction Machinery Audio Analysis: A GAN-Based Synthesis Method

Abstract

This study tackles the issue of limited construction machinery audio data, which restricts the development of early warning systems for protecting underground utilities during road construction. We propose a novel method for synthesizing audio data using GANs. By combining HIFI-GAN and BiLSTM networks, we develop a specialized model (H-GB) for audio synthesis. The model achieves a stable loss of 0.21, demonstrating high performance. Our experiments confirm the successful creation of high-fidelity audio data. This work enhances synthesized audio quality and the model's ability to learn construction-related audio features, providing a strong basis for advancing construction monitoring applications.