Oversampling with a K-fold increase in the rate of sampling digital audio files was studied. To save the sound frequency characteristics, the transition to a higher frequency should be accompanied by the insertion of К — 1 additional samples between each two adjacent original samples. The additional samples were generated by interpolation with two different splines — quasihermitian cubic (QHC-spline) and linear ones. At the end of the QHC-spline link central difference derivatives were calculated from the source file samples instead of the genuine ones used to construct hermitian splines. If the smoothness level is denned as the maximum of inverted second derivative modulo in the points of its continuity, it is proved that the oversampling with the sampling frequency increasing up by К times and with linear interpolation reduces the output signal smoothness level .ЙГ-fold in the neighborhood of the original node, but the oversampling by QHC-spline would drop the smoothness level of the recovered signal no more than 3 times for all K.