2518-1092Research result. Information technologies2518-109210.18413/2518-1092-2022-8-3-0-53225ARTIFICIAL INTELLIGENCE AND DECISION MAKING<strong>CLASSIFICATION OF SPEECH DATA BY EMOTIONAL BACKGROUND</strong><strong>CLASSIFICATION OF SPEECH DATA BY EMOTIONAL BACKGROUND</strong>ZhikharevAlexander GennadievichZhikharevAlexander Gennadievichzhikharev@bsu.edu.ruChernykhVladimir SergeevichChernykhVladimir Sergeevich20238300

In this paper, the algorithm of classification of speech data by emotional background, developed by the authors, is considered. In particular, it describes a neural network created to recognize eight different emotions in speech. To train the neural network, a training sample obtained from the RAVDESS dataset, which contains 1440 audio files, was used. These audio files contain the speech of 24 actors (12 women and 12 men) with a neutral North American accent. The paper describes the process of training a neural network using the Keras library, including the network architecture, layer sizes, activation functions and optimization methods. The stages of preprocessing and preparation of the original audio data before training the network are also discussed. The results of the study show that the developed neural network has high performance and the ability to recognize emotions with an accuracy of 80%.

In this paper, the algorithm of classification of speech data by emotional background, developed by the authors, is considered. In particular, it describes a neural network created to recognize eight different emotions in speech. To train the neural network, a training sample obtained from the RAVDESS dataset, which contains 1440 audio files, was used. These audio files contain the speech of 24 actors (12 women and 12 men) with a neutral North American accent. The paper describes the process of training a neural network using the Keras library, including the network architecture, layer sizes, activation functions and optimization methods. The stages of preprocessing and preparation of the original audio data before training the network are also discussed. The results of the study show that the developed neural network has high performance and the ability to recognize emotions with an accuracy of 80%.

audio attributesaudioaudio fileaudio dataemotional backgroundclassificationmodellayeraudio attributesaudioaudio fileaudio dataemotional backgroundclassificationmodellayerСписок литературыChollet F. Deep Learning with Python. 2nd interd. edition. &ndash; SPb.: St. Petersburg: St. Petersburg. &ndash; 576&nbsp;p.&nbsp;&ndash; ISBN 978-5-4461-1909-7.Han K., Lee K., Kim H.G. Music emotion recognition using chroma feature-based probabilistic neural network. Multimedia Tools and Applications. &ndash; 2017. &ndash; V. №76, Issue №3. &ndash; P. 3691-3710.Getting to Know the Mel-Spectrogram. [Electronic resource] &ndash; Electronic data, 2019. &ndash; URL: https://towardsdatascience.com/getting-to-know-the-mel-spectrogram-31bca3e2d9d0.Graves A., Schmidhuber J. Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Nature. &ndash; 2005. &ndash; V. №18, Issue №5-6. &ndash; P. 602-610.Understanding LSTM Networks. [Electronic resource] &ndash; Electronic data, 2015. &ndash; URL: http://colah.github.io/posts/2015-08-Understanding-LSTMs/.