Automation of the Spoken Poetry Rhyming Game in Persian

This paper aims to investigate how a Persian spoken poetry game, called Mosha'ere, can be computerized by using a Persian automatic speech recognition system trained with read speech. To do this, the text and recitation speech of the poetries of the great poets, Hafez and Sa'di, were gathered. A spoken poetry rhyming game called Chakame, was developed. It utilizes a context-dependent tri-phone HMM acoustic modeling trained by Persian read speech with normal speed to recognize beyts, i.e., lines of verses, spoken by a human user. Chakame was evaluated against two kinds of recitation speech: 100 beyts recited formally at the normal rate and another 100 beyts recited emotionally hyperarticulated at a slow rate. About 23% difference in WER shows the impact of the intrinsic features of emotional recitation speech of verses on recognition rate. However, an overall beyt recognition rate of 98.5% was obtained for Chekame.