Розробка ERP-застосунку для діаризації мовленнєвих команд

Abstract

The authors have developed a console application for diarization of speech signals in Python. The application is based on the previously created general scheme of voice biometrics GMM + i-vec + DNN. The accuracy of diarrhea for our application in a sample of 80 samples from 40 different speakers was 93%. In terms of speed, the average duration of voice processing when learning the system was 22 seconds (processed a file with a pronunciation duration of 20 seconds).