The default NFFT size of 256 gives misleading results if x.size << 256, for example this chirp:

import numpy as np
import pylab as pyl
from scipy.signal import chirp

fs = 200
t = np.arange(0, 1, 1/fs)

signal = chirp(t, t1=1, f0=0, f1=100)

pyl.specgram(signal, Fs=fs)
pyl.show()

Gives this. Passing NFFT=32, noverlap=16 to specgram() gives this more reasonable result.

Simplest fix is probably just to warn users when their data is too small.