Single channel speech enhancement with residual learning and recurrent network