Identification of Question and Non-Question Segments in Arabic Monologue Based on Prosodic Features Using Type-2 Fuzzy Logic Systems

• Main Authors: Olatunji, S O, Cheded, L, Al-Khatib, W G
• Format: Conference Proceeding
• Language: English
• Subjects: Arabic Audio Monologues; Content Analysis; Fuzzy logic; Fuzzy sets; Fuzzy systems; Prosodic Analysis; Question/Answer discrimination; Speech; Streaming media; Support vector machines; Type-2 Fuzzy Logic; Uncertainty
• ISSN: 2166-8523; 2166-8531
• DOI: 10.1109/CIMSiM.2010.41

In this work, we propose the use of type-2 fuzzy logic systems (type-2 FLS) to identify question and non-question segments in an Arabic monologue based on prosodic features. Prosody has been widely used in many speech-related applications including speaker and word recognition, emotion and accent identification, topic and sentence segmentation, and text-to-speech applications. An important application investigated is that of identifying question sentences in Arabic Monologue Lectures. Languages, other than Arabic, have received a lot of attention in this regard, hence the need for this research work concentrating on the Arabic language. Having first segmented the sentences from the continuous speech using energy and duration features, prosodic features are, then, extracted from each sentence. These features are used as input to the two proposed classifiers to classify each sentence into either Question or Non Question sentence. Results from this work have been compared with the previously used support vector machine (SVM), and the outputs indicate that the proposed type-2 FLS model outperforms SVM for all the experiments carried out, mainly due to its superior ability to handle uncertainties in the feature set.