Frequency-and-phase correction is an important step in the processing of single-voxel magnetic resonance spectroscopy data, and is required for J-difference editing, which relies on subtraction to reveal a low-SNR signal. We investigated an approach for frequency-and-phase correction using deep learning. Our networks were trained using simulated spectra manipulated with different frequency-and-phase offsets. During validation, the network returned spectra that were corrected to within 1.76 ± 1.19 degrees of phase and 0.09 ± 0.05 Hz of frequency, giving a difference spectrum very similar to the true unmanipulated spectrum. Frequency-and-phase correction is a promising application for deep learning in in vivo MRS.