A source signal (e.g. a voice test) is processed by a voice coder (1) or transmitted and converted into a receive signal (coded voice signal). The source and receive signals are separately subjected to preprocessing (2) and to psycho-acoustic modeling (3), followed by interval calculation (4) to evaluate the similarity of the signals. Finally, MOS calculation is carried out to obtain a result that can be compared with human evaluation. According to the invention, a spectral similarity value is determined for evaluating transmission quality, which is based on calculating the covariance of the spectra source and receive signals and dividing the covariance by the standard deviation of the above-mentioned spectra. The inventive method enables objective evaluation (voice quality forecast) while taking into account the human hearing process.