Emotion Recognition Audio Database for Service Robots

Abstract

This paper describes the first version of the emotional audio base for a service robot in which the sentences are in Romanian. The audio database is based on 10 sentences. Each sentence was recorded in three types of emotional states: happiness, neutral and sad, each recording being repeated 10 times. The meaning of each sentence also depends on its type of emotion. For creating the audio database, 7 people were recorded, including both female and male. For each person, 10 sentences were recorded. At this moment the database consists of 2100 audio sentences with various states of emotion. After obtaining this database, the next step would be to implement a sentence-based emotion identification algorithm for the use of service robots in various monitoring applications. In each emotional state, the average fundamental frequency for was calculated for both male and female speakers. The results indicate a significant impact of the happy emotional state upon frequency variation. The distinction between a neutral and sad emotional state is not particularly significant.