Simplicity of LSTM over Wavenet

[JZZQuant]

Hi,
I have observed that the LSTM based model is particularly very simple in comparison to the wavenet making it a very ideal candidate for real time applications. Both the convolution filters are linear with no non linear activation and even the receptive filed seems to be very short 100-150 unlike the wave-net where it is close to 1024. Does any one know the comparative results of LSTMs fair are they too biased due to their simplicity if not I am inclined to believe they are more efficient solution in this particular case especially with real time considerations in Mind