Error: Each initial covariance matrix must be symmetric and positive semidefinite.?

Hi, I am trying to fix, a guassian mixture model to the data using the command.

    obj = gmdistribution.fit(data,maxk,'Start',S,'CovType','diagonal','Options',opts, 'Regularize', 0.01);

I know the error is because the diagonal values in the covariance matrix are zero, But I use a regularize option to avoid it, but it still gives me that error. Is there any other way around this error?