Presents new and improved method of marking speech signals. The method is based on the use condition Gelder for the speech signal spectrum. Features of the method are its accuracy during decoding of the watermark and independence from the source data.