New version of NVM algorithm in case of stationary voice model are proposed for precise estimation of the Fundamental frequency on a short time interval. Its computational complexity is proportional to that of FFT on the same time interval. A precise trade-off between approximation error and numerical speed is established.