SYN-N & REAL-N: Negative Acoustic Dataset for Blast Sound Discrimination

The negative dataset comprises two subsets: SYN-N and REAL-N (0.5s duration, 48kHz sampling rate). The SYN-N subset contains simulated transient pulses with characteristic periods T_ID strictly outside the target distribution (less than 0.8 ms or greater than 4.0 ms). The REAL-N subset provides a diverse collection of real-world audio, including system noise, environmental sounds, and non-Friedlander battlefield acoustics. Designed as "hard negatives," this combined dataset is crucial for training deep learning models to enhance noise robustness and minimize false alarm rates in complex, real-world environments.