Densely-connected Convolutional Recurrent Network for Fundamental Frequency Estimation in Noisy Speech