Development and Research of Neural Network Methods for Recognizing Noisy Speech Audio Files

Abstract

In this work an approach to training neural networks for speech transcription using artificially generated and noisy data is investigated and developed. The purpose of the work is to determine the effectiveness of the proposed methods for recognizing noisy speech using neural network technologies. The relevance of the study is determined by the emerging new possibilities for noise compensation in audio files using neural network technologies. A study of existing noise affecting the training of a neural network was carried out. A data noise module was designed and developed, and a test bench was assembled. In the course of practical research, a slight improvement in text recognition by a neural network for speech transcription was identified, and options for improving the quality of recognition were proposed. Methods of system analysis and experimental research determine the methodology of this study.