Универсалии и специфика выражения эмоций в балкарском языке (исследование на основе алгоритмов машинного обучения)
Universals and specifics of emotion expression in the Balkar language (research based on algorithms machine learning)

Summary/Abstract: The paper aims to examine of the universal and specific features of emotion expression in the Balkar language and to provide an efficient and accurate technique for Balkar speech emotion recognition. Specifically, we present PySound as a tool to transcribe and analyze phonetic data which derives prosody attributes from emotional speech as extra features to improve emotion recognition as conventional automatic emotion recognition systems mostly rely on spectral features which are greatly affected by outer factors. Regarding the methodology, we gathered statements from representatives of Russian and Balkarian ethnic groups, marked with emotional states of “joy” and neutral versions of the same dialogues. The informants were females aged 25—45 years, who do not live in rural areas, speak the Balkar language and use it in everyday communication. The acoustic analysis was conducted based on the intensity of phonetic syllables and the fundamental frequency (F0), both normalized using Lobanov’s z-score. The study found that the ‘emotion-specific’ speech features in the Balkar language is associated with a more active role of the fundamental frequency in the syllable prominence and a phrase’s different dynamic organization compared to the Russian language. The experiment results show that combining prosody and MFCC features yields an overall accuracy of 74 % for emotion recognition, which makes improvement compared with using the single prosody or MFCC features. The prosodic features’ contribution analysis to the model’s performance showed that tonal characteristics play a relevant role, accounting for around 30 % of the total feature contribution

• Details
• Contents