Evaluation of the Quality of Compression of Acoustic Signals in the Low-Frequency and Mid-Frequency Sound Range Using a Neural Network Codec

This article compares the compression algorithm using a neural network codec with modern audio compression algorithms such as MP3, AAC, WMA, ALAC, FLAC. The results of the compression quality evaluation taking into account various parameters of signals before and after compression are presented, as well as a comparative characteristic of the algorithm execution speed on AMD Ryzen 7 5800X, Intel(R) Core(TM) i3-5005U, AMD Ryzen 5 3600, AMD Ryzen 5 2500U, Intel(R) Core(TM) i5 processors-10400F. Experiments have proved that the use of a neural network codec is possible in real-time systems on a single processor core with low performance, but the speed of the neural network codec is many times lower than that of classical algorithms. Also, it was revealed that the compression quality of the neural network codec can compete with lossless compression algorithms.