COVID-19 pandemic has sent ripple effects across the world affecting various economic and social parameters. The spread of the pandemic has been observed to be dynamic and the various measures implemented in different countries had been largely based on the guidelines of maintaining hygiene, frequent disinfection, social distance norms and prevention of community spread. There is an observable trend embedded with stochastic oscillations in pandemic's progress in every country. The fundamental similarity in the pandemic's progress in countries can be deciphered using clustering and this similarity can be used to model the future dynamics of countries from those clustered together. The time-frequency spectral estimation gives the temporal dynamics, frequency and corresponding energy deposited by the epidemic. This work uses Wigner energy spectrum of the pandemic to derive spectral parameters namely energy membership probabilities, deviation from maximum spectral energy for all temporal and frequency deposits and the maximum energy spectral deposit as latent variables with EM algorithm for clustering different countries. The modified EM algorithm with two layer latent representations effectively captures the similarity in the temporal dynamics of different countries.