Kannada Emotional Speech Database: Design, Development and Evaluation
Emotion is a state of cognizant that involves sentiment and plays an important role in communication. This paper illustrates the development of Kannada Emotional Speech (KES) database and the process of its evaluation. The purpose for developing the datab
- PDF / 220,704 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 57 Downloads / 184 Views
Abstract Emotion is a state of cognizant that involves sentiment and plays an important role in communication. This paper illustrates the development of Kannada Emotional Speech (KES) database and the process of its evaluation. The purpose for developing the database is for analysis of acoustic features, building an effective emotion conversion and emotion recognition system for human machine interaction. The KES database consists of acted emotional sentences in regional Kannada Language. The evaluation of database is performed using Mean Opinion Score (MOS), K-NN (K-Nearest Neighbour) and LVQ (Learning Vector Quantization) classifiers. The evaluation of basic emotions (sadness, happy, fear and anger) as well as neutral was carried out on speech samples of adult speaker (male/female) and child speaker (male/female). The emotions were recognized well above acceptance level for all the speakers. Keywords KES
MOS K-NN LVQ
1 Introduction Emotional Speech analysis is an important field of research in Computer science. One of the key aspects for natural ness in speech is producing appropriate emotional expression. Adding emotion to the synthetic speech will decrease the monotony and improves human machine communication. In past many research has been done on speech databases for several languages, speech analysis (macroscopic or in phoneme by phoneme basis) and prosody modelling [1–3]. A complete literature A. Geethashree (&) D.J. Ravi Vidyavardhaka College of Engineering, Mysore, India e-mail: [email protected] D.J. Ravi e-mail: [email protected] A. Geethashree D.J. Ravi Visvesvaraya Technological University, Belagavi, India © Springer Nature Singapore Pte Ltd. 2018 D.S. Guru et al. (eds.), Proceedings of International Conference on Cognition and Recognition, Lecture Notes in Networks and Systems 14, DOI 10.1007/978-981-10-5146-3_14
135
136
A. Geethashree and D.J. Ravi
review on qualitative analysis of emotional speech is found in [4]. Analysis of emotional speech is an important aspect in emotion conversion and emotion recognition system. For this purpose a proper emotional database is required. The language that we speak not only carry linguistic features but also conveys nonlinguistic features, such as speaker’s emotion, gender, social status, age etc. India being a multilingual country, studies in the field of emotion recognition and emotion conversions has been done in English, Hindi and other languages [5–7]. There is also the need to study the emotional aspects in Kannada speech. The KES database is the first one developed in Kannada for analyzing the emotions present in speech. This study would provide speech language pathologist and understanding on the normal and abnormal aspects of prosody parameters, which would help them to analyze the individual’s with communication disorders. The prosodic features (Pitch, energy and duration) are important for the production of each emotion [8, 9]. There is also a correlation between emotion and glottal features (pitch contours, prosodic units accents
Data Loading...
