For example,Бобцов

APPROACH TO AUTOMATIC RECOGNITION OF EMOTIONS IN SPEECH TRANSCRIPTIONS

Annotation

The issue of recognizing emotions in speech transcriptions, which is relevant in various fields, is studied. The influence of preprocessing methods (stop word removal, lemmatization, stemming) on the accuracy of emotion recognition in text data in Russian and English is analyzed. To conduct experimental studies, orthographic transcriptions of dialogues from the multimodal corpora RAMAS and CMU-MOSEI in Russian and English, respectively, are used. These corpora are annotated for the following emotions: joy, surprise, fear, anger, sadness, disgust and neutral. Preprocessing of text data includes removal of punctuation marks and stop words, tokenization, lemmatization and stemming. Vectorization of the resulting material is carried out using the TF-IDF, BoW, Word2Vec methods. The used classifiers are support vector machines and logistic regression. An approach is developed that is a combination of the above methods. For the Russian language, the highest accuracy of emotion recognition achieved using a weighted F-measure is 92.63 %, for the English language – 47.21 %. In addition, studies are conducted to identify the number of remote stops for effective emotion recognition from text data. Experimental results show that storing stop words in the source text allows to achieve the highest accuracy of text classification.

Keywords

Постоянный URL

Articles in current issue

CONTENT © ITMO UNIVERSITY 1993-2024