Acoustic Emotion Recognition for Improved Safety in the Construction Industry

In the construction industry, clear communication is vital for optimal teamwork and performance. Yet, diverse languages, accents, and continuous worksite noises often impede speech clarity, leading to misunderstandings and consequential errors. Addressing this challenge, the presented research introduces a speech emotion recognition algorithm tailored for construction. Unlike prior algorithms emphasizing language content, this approach centers on speech features, bridging a significant gap in construction-specific applications. The algorithm aims to identify four key emotions: anger, happiness, sadness, and neutrality, using six distinct acoustic features: pitch, intensity, frequency formants, jitter, shimmer, and zero crossing rate. Leveraging these features collectively enhances system accuracy. Designed in MATLAB, the decision-tree-based method calculates confidentiality intervals for each feature and is specifically developed for speaker-dependent emotion recognition. Remarkably, the algorithm secured an 86% accuracy rate in detecting anger, outperforming existing models. By recognizing and addressing the emotional states of workers, this method holds the promise to greatly enhance safety on construction sites, averting potential hazards.