Can you provide a minimal example to demonstrate the problem and the image produced?

Your example will probably be turned into a test so this problem does not come back.

I am sorry, I tried really hard to build a simple code but it all worked fine.
The problem only happens when using data from https://www.kaggle.com/c/seizure-prediction/data
so there is something strange in these specific matlab files.
Anyway, once you have downloaded one of the files, for example Dog_1.tar and open it, you can run the following code:

from matplotlib import pylab as pl
import scipy.io

data = scipy.io.loadmat('Dog_1/Dog_1_interictal_segment_0019.mat')
d = data['interictal_segment_19']['data'][0,0]
pl.specgram(d[0,:])
pl.show()

and 9 horizontal lines appear in equal distance (splitting the y axis to 8 segments)
but I found that if you copy the data everything looks good:

pl.specgram(d[0,:].copy())
pl.show()

which hint that the problem is somehow related to numpy view mechanism

Can you try with 1.4.x branch? This looks like (unable to test at the moment to verify) the strided view bug (#3467) which will be fixed in 1.4.1

Yes the problem is solved in 1.4.x and 1.5.x
(and a work around for 1.4.0 is to copy the input before calling specgram)

Thanks