Abstract :
This paper reports a compilation of a corpus of speech segmented into illocutionary performance units which are annotated with illocutionary forces with associated emotions and prosodic features. The speech data are sampled from two sources: (1) from the traditional story-telling broadcast over the radio; (2) from recordings of natural occurring talk activities. The first represents voice-acting data, in which there is a loss of naturalness, but is compensated with advantages of high resolution, low noise, and clearly articulated emotions. The second is praise-worthy of being realistic, reflecting real-life situations, but suffers from low resolution, a lot of noise, and mixed emotions, which resist categorical differentiation.
