Coded Speech Enhancement Using Neural Network-Based Vector-Quantized Residual Features