SITB-OSED: An Odia Speech Emotion Dataset

This Odia is one of the old Eastern Indo-Aryan languages, spoken by 49.8 million people across India. This article delineates the first open-sourced large speech emotional database in Odia language, named SITB-OSED. A key hurdle for effective speech emotion recognition (SER) systems is the lack of large corpora annotated with different emotional labels. This SER database is presented along with the consideration of its reliability based on machine learning analysis. The database contains six basic emotions, recorded in a quiet environment with the corresponding emotional texts by professional speakers. Speech data is of great interest to all research groups working on speech analysis and automatic speech recognition. Our database currently contains 5 major Odia dialects of speech data with a total of 12,110 utterances. We also tested several SER baseline systems on the SITB-OSED dataset and show the impact of recording diversity on performance. The corpus can be useful for both academic research and practical application, and will be available on https://www.speal.org/