Methods and arrangements for generating a voice model in speech processing. The method of the invention is characterized in that when determining a simulated signal Ssim, specific global reflection coefficients and receiver-receiver functions are introduced in feedback loops. The side model may be trained jointly with the training of the quantizer and is configured to model low priority data among the input data. Also, the article includes an underfoot strand that is coupled to the longitudinal strand and that extends across the sole structure to extend between the lateral side and the medial side of the upper.