As the $W$ matrix entries follow a multivariate normal distribution, then the probability that this matrix is singular is zero. 

That makes the matrix full rank, implying that $W^\dagger = W^{-1}$. That makes the last distance be zero : 
$\mid \mu- WW^\dagger\mu\mid = \mid \mu - I\mu\mid = 0$