
Investigation of the quality of audio encoding by different encoders

The MP3 format for high-quality audio encoding is becoming increasingly popular. Originally developed for use within the MPEG1 and MPEG2 video compression standards, it quickly became mainstream as a standalone format. The main reasons for this turn of events were the preservation of high sound quality with high compression ratios and the more than active attempts by the developer IIS Fraunhofer to make the most of his creation.

In essence, MP3 is a direct evolution of MPEG Layer I and Layer II, and it also uses a psychoacoustic model to encode the original signal. Because of this, the encoding process is ambiguous and may vary depending on the encoder used (for more details, see MP3 Overview. Part 1). This ambiguity means that, having encoded the same signal with two different encoders, we can obtain, after decoding, two different sound signals. Obviously, the preferred encoder is the one that best preserves the original signal. The purpose of this review is to find out which modern encoder will give us the best result.
Test methodology
Generally, to compare the original and encoded signal, the method of comparing its amplitude frequency characteristics (AFC) is used. There are two varieties of this method: comparing the average frequency response of the signals, and comparing the change in frequency response over time. The first type is used most often due to the simplicity of its implementation: the comparison needs to be done only once. However, during averaging, a significant part of the information about the signal is lost, and as a consequence, with an absolutely identical frequency response, the original and encoded signal can differ greatly in sound if the signal contains a large amplitude. , but very short. bursts at term of some frequencies. The second type allows you to avoid such problems,
Not so long ago, a method for comparing sonograms of signals became widespread: a graphical representation of the frequency response of signals over time, in which time is plotted along the abscissa, the frequency of the component of the signal is plotted along the ordinate (generally a logarithmic scale is used), and the intensity of the luminescence of the points determines the amplitude of this frequency component of the signal. This method is essentially a modification of the method for comparing changes in frequency response over time, and the problem of underperformance is solved by reducing the number of frequency components considered and expanding the “scanning window” of the frequency response at 50-100. samples, allowing you to use FFT. This simplification of the method inevitably leads to a decrease in its precision. First, a decrease in the number of frequency components considered leads to a loss of the “resolution” of this method, making it practically analogous to averaging the frequency response over time. Second, due to the magnification of the scanned window and the use of the FFT, there is the effect of “smearing” the signal in time.














