The MP3 Encoder is that program that analyzes the uncompressed digital file (for example, a Wav file) and transforms it into an MP3 file.
The audio signal is filtered and divided into 576 areas (called subbands) through a process that uses DCT (Discrete Cosine Transformation) and manages to eliminate all unnecessary frequencies. The human ear, as already stated, perceives sounds only beyond a certain threshold so that all the audio below is not encoded.
At this point, the resulting signal passes through the psychoacoustic model in which the masking thresholds of which we have spoken previously are identified. This is done using the discrete Fourier transform (DFT, Discrete Fourier Transform).
During the masking of the 576 subbands, the frequencies to be masked are determined and therefore can be removed.
After masking, the defined Stereo Ensemble process is applied. Below a certain frequency, the ear cannot perceive the spatial position of sounds, so they can be recorded on a single channel (therefore in mono format) with significant space savings.
Once the file is ready, the data is further analyzed and compressed using Hufmann encoding which allows for a data reduction (without loss of information) of approximately 20%.
At this point, after all the data has been collected, the encoder proceeds to create the bit stream that will form the final MP3 file.
Compression criteria
To perform such compression, the MP3 format is based on a simple concept: filter a digital musical piece and eliminate all unnecessary information, thus reducing space.
The human ear is an almost perfect instrument but it also has its limits. The human ear pass band extends from 20 Hz to 20,000 Hz, but is much more sensitive to those in the mid-range, 700 to 6,000 Hz, where most of the information is concentrated.
The study of auditory perception is a matter of psychoacoustics that mainly analyzes 2 factors that are later used in MP3 encoding:
Auditory perception
In the area of sounds, only a few can be heard by the human ear. The following figure shows these areas that represent the different sound frequencies. Only those in the white area are audible from our ear.
Masking
Masking is nothing more than the superposition of weak sounds with loud sounds. It almost always happens that the sounds of different instruments overlap each other. In cases where the loudest sound completely covers the lowest, there is a so-called masking. In MP3 files, masking allows you to remove the information from the weakest sounds, which, however, because they are not perceived by the ear, are virtually irrelevant.
