Open-Office Noise Level Classification Dataset

As modern office environments become more open and flexible, noise pollution has increasingly become a major hindrance to productivity in workplaces. Currently, noise generated by conversations, equipment operations, and other sources in most open offices severely affects employee concentration and work efficiency. However, existing noise control solutions often fail to provide accurate and effective solutions due to the lack of data identifying noise sources and types specific to the office environment. The construction of this dataset aims to enhance the precision and effectiveness of noise management by identifying and classifying common noise types in open offices. The audio data in this dataset is collected using high-sensitivity recording equipment placed at different locations within the open office to capture diverse noise sources. The data undergoes multiple rounds of annotation to ensure accuracy and consistency of classification labels, supervised by a professional team with an acoustic background. The data preprocessing steps include noise filtering, audio clipping, and format conversion, with final storage in WAV format organized by classification labels. The core strengths of the dataset are its high quality and diversity, with annotation accuracy reaching over 95%, ensuring data consistency and completeness. New annotation methods and data augmentation techniques, such as speed change playback and frequency transformation, are introduced to effectively improve the robustness of models. By integrating various noise scenarios, it significantly improves noise management in office environments. Compared to similar datasets, its diversity and fine-grained annotations are competitive advantages. The dataset is not only suitable for various office environments but also extends to noise analysis in other noisy environments, offering broad versatility.