This work proposes a methodology to improve the extraction of coherent structures associated with the generation of acoustic fluctuations in turbulent jets from high-speed Schlieren images. This methodology employs the momentum potential theory of Doak [14,24] to compute potential (acoustic and thermal) energy fluctuations from the Schlieren images by solving a Poisson equation, in the manner introduced by Prasad & Gaitonde [38]. The calculation of momentum potential fluctuations is then combined with the spectral proper orthogonal decomposition (SPOD) technique: the cross-spectral density is defined based on the momentum potential field, instead of the Schlieren images. While the latter are dominated by a broad range of vortical fluctuations in the turbulent mixing region of unheated high-speed jets, the momentum potential field is governed by acoustic fluctuations and its spatial structure in the frequency domain is remarkably coherent. This approach is applied here to Schlieren visualizations of a twin-jet configuration with a small jet separation and two supersonic operation conditions: a perfectly-expanded and a overexpanded one. The SPOD modes based on momentum potential fluctuations retain the wavepacket structure including the direct Mach-wave radiation together with upstream-and downstream-traveling acoustic waves, similar to SPOD modes based on the Schlieren images. However, they result in a remarkably lower-rank decomposition than Schlieren-based SPOD and, as opposed to the latter, provide an effective separation of twin-jet fluctuations into independent toroidal and flapping oscillations that are recovered as different SPOD modes.