Detection of question segment in an audio Arabic lecture

Abstract

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 we investigated, is that of identifying question sentences in Arabic monologue lectures. Languages other than Arabic have received a lot of attention in this regard. We approached this problem by first segmenting the sentences from the continuous speech using intensity and duration features. Prosodic features are then extracted from each sentence. These features are used as input to decision trees to classify each sentence into either Question or Non-Question sentence. Our results suggest that questions are redundantly marked in natural Arabic speech and automatically extracted prosodic features can make significant contribution in question identification. We classified Questions with an accuracy of 77.43%. Feature specific analysis further reveals that energy and fundamental frequency (F0) features are mainly responsible for discriminating between question and non-question sentences. We found that Bayes Network performed better than SVM, MLP and Decision Trees on our dataset. Removal of correlated features through Correlations Based Feature Selection (CFS) produced more efficient and accurate results than the complete feature set.