A comparison between spiking and differentiable recurrent neural networks on spoken digit recognition