In natural conversations, significant part of laughter co-occurs with speech which is referred to as speech-laugh. Hence, speech-laugh will have characteristics of both laughter and neutral speech. But it is not clearly evident how acoustic properties of neutral speech are influenced by its co-occurring laughter. The objective of this study is to analyze the acoustic variations between vowel regions of laughter, speech-laugh and neutral speech. The features based on excitation source characteristics extracted at epochs are considered in this study. Features extracted in the vowel regions of speech-laugh exhibit deviations from that of laughter and neutral speech. These deviations in feature values are exploited to discriminate speech-laugh from laughter and neutral speech. Two different datasets consisting of conversational speech and meeting recordings are used in this analysis. Experimental results show that the discrimination between the three classes obtained by considering vowel regions is better than that of considering the complete utterance.
展开▼
