Audio Dataset for Abnormal Sound Extraction in Open Office Environment

In open office environments, detecting abnormal sounds is crucial for enhancing security and work efficiency. However, the complex acoustic environment and diverse sound types make traditional monitoring solutions ineffective. Existing solutions have limitations in accuracy and real-time capability, making it difficult to handle sudden and abnormal acoustic events. This dataset aims to address the needs for sound information extraction accuracy and real-time monitoring. Data is collected using directional microphones in real office environments, covering various sudden sounds during working hours. Quality control involves multiple rounds of manual annotation and machine verification to ensure data consistency and high precision. The annotation team consists of experienced acoustic engineers and data scientists, with a total scale of over 20 people. Data preprocessing utilizes noise reduction filters and feature enhancement techniques. Audio data is stored in WAV format and organized according to sound event types.