Frames without padding
'pre_processing.make_frames' now takes arg 'zero_padding' with default value set to False (previously True).
This solves the problem with end-point artifacts in spectrograms.
Note also that the AudioSignal class now recasts the signal as np.float32