In this paper, we are introducing a database for speaker recognition in multilingual and emotional conditions. The paper describes the design, collection and preliminary investigations carried out on a multilingual emotional speech database, named Arunachali Multilingual Emotional Database (AMLED), developed using English, Hindi and the languages of Arunachal Pradesh, namely, Adi, Nyishi, Galo, Apatani and Monpa, spoken by the natives of Arunachal Pradesh, a north east frontier state of India. The speech corpus consists of two different categories of emotions– simulated or acted and emulated or induced. The speech corpus has been recorded from 100 native speakers of Arunachal Pradesh. Each speaker has recorded for four different types of emotions– angry, happy, sad and neutral. The neutral voice of each speaker is recorded in three different languages– English, Hindi and the mother tongue of the speaker, which is any one of the above five Aruanchali languages. The proposed database will be useful for speaker verification at emotional and multilingual conditions. Further, the database can be used for identifying acoustic cues that distinguish true emotions from the acted emotions. To determine the quality of the emotions present in the speech signals, subjective listening test has been conducted. Statistical analysis of the features extracted from the speech signals recorded at different emotional and linguistic conditions have been carried out to determine the changes at parameter level with change in emotional and linguistic conditions.

Emotional Speech Corpus; Multilingual Speech Corpus; Simulated Emotion; Emulated Emotion; Subjective Listening Test; Statistical Analysis.