Excess memory usage in kmeans training

When training on voxforge and 256 GMMs with dask partition size of 200, I get this error:

bob.learn.em/bob/learn/em/k_means.py", line 78, in e_step
    np.eye(n_clusters)[closest_k_indices][:, :, None] * data[:, None],
numpy.core._exceptions.MemoryError: Unable to allocate 7.39 GiB for an array with shape (64546, 256, 60) and data type float64