A Dataset for Speech Emotion Recognition in Greek Theatrical Plays

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1040–1046 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1040 A Dataset for Speech Emotion Recognition in Greek Theatrical Plays Maria Moutti1, Sofia Eleftheriou2, Panagiotis Koromilas2, Theodoros Giannakopoulos2 3 1University of the Peloponnese, 2National Center for Scientific Research - Demokritos, 3Behavioral Signal Technologies Inc. {mar.moutti}@gmail.com, {seleftheriou, pakoromilas, tyianak}@iit.demokritos.gr Abstract Machine learning methodologies can be adopted in cultural applications and propose new ways to distribute or even present the cultural content to the public. For instance, speech analytics can be adopted to automatically generate subtitles in theatrical plays, in order to (among other purposes) help people with hearing loss. Apart from a typical speech-to-text transcription with Automatic Speech Recognition (ASR), Speech Emotion Recognition (SER) can be used to automatically predict the underlying emotional content of speech dialogues in theatrical plays, and thus to provide a deeper understanding of how the actors utter their lines. However, real-world datasets from theatrical plays are not available in the literature. In this work we present GreThE, the Greek Theatrical Emotion dataset, a new publicly available data collection for speech emotion recognition in Greek theatrical plays. The dataset contains utterances from various actors and plays, along with respective valence and arousal annotations. Towards this end, multiple annotators have been asked to provide their input for each speech recording and inter-annotator agreement is taken into account in the final ground truth generation. In addition, we discuss the results of some indicative experiments that have been conducted with machine and deep learning frameworks, using the dataset, along with some widely used databases in the field of speech emotion recognition. Keywords: GreThE dataset, speech emotion recognition, Greek theatrical plays, valence, arousal 1. Introduction The task of recognizing the underlying emotion from speech, irrespective of its semantic content, is rather important in various applications. However, it is hard to notate even by human beings, mostly due to the subjectiveness of the emotional content. The ability to automatically conduct it, is a demanding task and still an ongoing subject of research. Several open- source databases exist in the field of Speech Emotion Recognition (SER) which may contain audio-only or multimodal information and are usually annotated on two categories: categorical attributes (distinct classes of emotions) (Ortony and Turner, 1990; Ekman, 1992) and dimensional attributes (continuous values of valence, arousal and intensity) (Wundt and Judd, 2011; Russell, 1980). The emerge of the fields of Machine and Deep Learning in the past decade, gave the chance to researchers to apply research outputs on real-world problems. However, despite the fact that SER is a task that has gained great attention in the literature (Koromilas and Giannakopoulos, 2021), the industrial applications of the proposed works are either centered around web content (online video or podcast analysis) or have been applied on actual conversations to enrich understanding (eg. empathetic dialogue (Ma et al., 2020)). At the same time, cultural events are a base factor for every human civilization and, …

Content truncated to 500 words