IIIT-H TEMD Semi-Natural Emotional Speech Database from Professional Actors and Non-Actors

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1538–1545 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1538 IIIT-H TEMD Semi-Natural Emotional Speech Database from Professional Actors and Non-Actors Banothu Rambabu, Botsa Kishore Kumar, Paidi Gangamohan, Suryakanth V Gangashetty Speech Processing Laboratory International Institute of Information Technology-Hyderabad, Telangana, India {rambabu.b, kishore.botsa, gangamohan.p}@research.iiit.ac.in, svg@iiit.ac.in Abstract A fundamental essence for emotional speech analysis towards emotion recognition is a good database. Database collected from natural scenarios consists of spontaneous emotions, but there are several issues in collection of such database. Other than the privacy and legal related concerns, there is no control over environment at the background. As it is difficult to collect data from natural scenarios, many research groups have collected data from semi-natural or designed procedures. In this paper, a new emotional speech database named IIIT-H TEMD (International Institute of Information Technology-Hyderabad Telugu Emotional Database) is collected using designed drama situations from actors and non-actors. Utterances are manually annotated using a hybrid strategy by giving the context to one of the listeners. As some of the data collection studies in the literature recommend for actors, analysis of actors data versus non-actors data is carried out for their significance. The total size of the dataset is about 5 hours, which makes it an useful resource for the emotional speech analysis. Keywords: Emotional speech, Natural database, Semi-natural database 1. Introduction Research on emotions (especially) in speech communica- tion has got more importance. This is due to emergence of technologies like smart phones, intelligent robots, spoken dialogue systems, and many human-machine speech based interactive systems. These technologies can be exploited effectively if they extract emotions from the speech sig- nal. One of the major limitations in developing emotional speech based systems is lack of good databases. Ideally, a database consisting of ‘spontaneous’ emotions collected from natural scenarios is required. But it is difficult to col- lect such a database, mainly due to privacy concerns and legal issues. Databases collected by several research groups can be broadly categorized into simulated parallel, semi-natural, and (near to) natural databases (Douglas-Cowie et al., 2003), (Koolagudi and Rao, 2012), (Schuller et al., 2011). Most of the simulated parallel and semi-natural databases were collected from speakers by asking them to elicit spe- cific emotions through scripted scenarios. Performance of emotion recognition systems developed on such databases degrades in real-life scenarios (Busso et al., 2008), (Camp- bell, 2003). There are many issues, such as selection of speakers, design of scenarios, recording environment, and recording equipment, in collecting a simulated emotional speech database close to real-life scenarios. Widely used simulated parallel databases such as IITKGP- SESC (in Telugu language)(Koolagudi et al., 2009) and German EMO-DB (Burkhardt et al., 2005) were collected through straight forward process, where the speakers were asked to emote a sentence in all emotions. The speakers in- volved in these recordings were professional voice talents. The major disadvantages of these kind of recordings are: 1) Speakers are asked to …

Content truncated to 500 words