Abstract:
The purpose of this work is to review the most successful open solutions in the field of speech
recognition and also considers the processes of speech recognition and the possibilities of their practical
use. The paper presents classical solutions based on recurrent neural networks, as well as more modern
ones, which use convolutional neural networks as a basis to remove noise and reduce dimensionality, and
transformers that allow to memorize the context and work with the semantic meaning of sequences,
regardless of time.