Development of Javanese Speech Emotion Database (Java-SED)

Abstract

Javanese is one of Indonesia's most widely spoken regional languages, alongside other regional languages. Emotions can be recognized in various ways, including facial expression, behavior, and speech. Recognizing emotions through speech is a straightforward process, but the outcomes are significant. Currently, there is no database for identifying emotions in Javanese speech. This paper aims to describe the creation of a Javanese emotional speech database. Actors from the Kamasetra UNY community who are accustomed to performing in dramatic roles participated in the recording. The location where recordings are made is free of interference and noise. The actors of Kamasetra have simulated six types of emotions: happy, sad, fear, angry, neutral, and surprised. The cast comprises ten people between 20 and 30 years old, including five men and five women. Both humans (30 Javanese-speaking verifiers ranging in age from 17 to 50) and a machine learning system (30 Javanese-speaking verifiers with ages between 17 and 50) verify the database that has been created. The verification results indicate that the database can be used for Javanese emotion recognition. The developed database is open-source and freely available to the research community at this link https://beais-uny.id/dataset/.