Neural Net as Pseudo-Inverse Filter in Speech Coding Problem

In the general case, a finite impulse response (FIR)filter has no inverse filter. To solve the inverse filtering problem, we propose an approximate method that restores the initial sequence. We analyze the sequence obtained by filtering the source sequence with an arbitrary FIR filter. The analysis made with the help of a deep neural network. The results of the study are applied to speech coding. We show experimentally, that our approach provides lowering bit rate in the transmission of speech data through a channel.