We applied the machine learning (ML) dipole moment models to calculate the dielectric properties of various liquid alcohols. This method assigns the Wannier centers (WCs) to chemical bonds between atoms and utilizes deep neural networks to predict the positions of these WCs for each bond. Our approach accurately reproduces the WCs of individual bonds and the dipole moments of entire molecules, validating the usefulness of the bond dipole scheme. We show that the polarization of WCs greatly enhances the dipole moment and dielectric constant in the liquids due to local intermolecular interactions. The calculated dielectric spectra quantitatively agree with experiments over terahertz (THz) to infrared regions. Our method applies to other molecular liquids and can be widely used to study their dielectric properties.