On the comparison of LES data-driven reduced order approaches for hydroacoustic analysis