<p>In this paper, we propose a class of shrinkage estimators for the mean parameter matrix of singular multivariate elliptically contoured distributions in the context of high- and ultra-high-dimensional data with unknown covariance. In particular, we generalize the existing methods in four aspects. First, we weaken the assumptions regarding the distribution and dimensions of the random sample. Second, we relax the assumption about the invertibility of the covariance, which represents a nuisance parameter. Third, we establish sufficient conditions under which the proposed class of shrinkage estimators has finite risk function. Fourth, under a very general loss function, we derive the superiority of the proposed shrinkage estimators over the classical sample mean. Furthermore, our results provide greater flexibility by addressing the challenges of high-dimensional data across a broad class of distributions, while also extending recent findings that were previously restricted to quadratic loss functions.</p>