The Use of Correlation Features in the Problem of Speech Recognition

The problem solved in the article is connected with increase efficiency of phraseological radio exchange message recognition, which sometimes takes place conditions increased tension for pilot. For high-quality signal preprocessing methods are needed. considers new data algorithms used to extract features from a speech message. In this case, two approaches were proposed. first approach building autocorrelation functions messages based on Fourier transform, second one uses idea portraits signals. proposed quite simple implement, although they require cyclic operators, since work pairs samples original signal. Approbation developed method was carried out recognizing Russian. algorithm preliminary feature extraction provides gain 1.7% recognition accuracy. use convolutional neural networks also an efficiency. processing about 3–4%. Quantization optimize models. algorithm’s performance by 2.8 times after quantization. It possible accuracy 1–2% using digital algorithms. An important possibility generalizing them arbitrary time correlation. discussed classical constructing series has novelty. At same time, ensures high However, study showed that all under consideration perform poorly influence strong noise.