How to compare two audio files quality wise?

February 28, 2009
89 Views

In order to compare two audio signals we introduce theanalytical module, which compares separately combined pairs of fragments ofactive and inactive phase signal that allows getting more accurate estimation.

For each fragment we determine integral spectrum by usingdiscrete cosine transformation (DCT). Spectrum integration is calculatedaccording to the proprietary formula. In the spectrum calculation the interpenetrationof windows comes to N/2 samples, and the Hamming or Blackmann-Harris windowfunction is applied to every window. Levels of spectrum energy on bands aredetermined for all sets of bands. Groups of critical bands, determined bydifferent authors resulting from different models of sound perception andspeech production.

Band boundaries (initial and terminal indices) as well asband energy values we determine by a set of proprietary formulas. The initialquality estimation value is taken as 100%, which decreases proportionally todistinction of energies on bands. The most interesting fact is that we canscale down our percentage of signals similarity to the well-known Mean OpinionScore (MOS) values, which correspond in tests to Cisco MOS or ITU P.862 asprecise as 97%. More intere

In order to compare two audio signals we introduce theanalytical module, which compares separately combined pairs of fragments ofactive and inactive phase signal that allows getting more accurate estimation.

For each fragment we determine integral spectrum by usingdiscrete cosine transformation (DCT). Spectrum integration is calculatedaccording to the proprietary formula. In the spectrum calculation the interpenetrationof windows comes to N/2 samples, and the Hamming or Blackmann-Harris windowfunction is applied to every window. Levels of spectrum energy on bands aredetermined for all sets of bands. Groups of critical bands, determined bydifferent authors resulting from different models of sound perception andspeech production.

Band boundaries (initial and terminal indices) as well asband energy values we determine by a set of proprietary formulas. The initialquality estimation value is taken as 100%, which decreases proportionally todistinction of energies on bands. The most interesting fact is that we canscale down our percentage of signals similarity to the well-known Mean OpinionScore (MOS) values, which correspond in tests to Cisco MOS or ITU P.862 asprecise as 97%. More interesting facts to follow – stay in touch!