Automatic Language Identification from Spectorgam Images

Abstract

The main idea of the work is to apply CNN and LSTM algorithms on the preprocessed audio data converted to spectrograms. In the experiment 7 languages are used English, Kazakh, French, German, Italian, Russian and Spanish to test that algorithm which show over 99% training accuracy and the maximum of 94.28% accuracy on the test set. In the end there is a discussion of the 100% of classification of the Kazakh language and the possible influences of dataset on the result.