Syllable detection in read and spontaneous speech

Abstract

Automatic syllable detection is an important t ask when analysing v ery large speech corpora in order to answer questions concerning prosody, r h ythm, speech r a t e, speech recognition and synthesis.In this paper a new method for automatic detection of syllable nuclei is presented.Two large spoken language corpora (PhonDatII, Verbmobil) were labelled by t hree phoneticians and t h en used to adjust the k ey parameters of the algorithm and to evaluate its error rate.Additionally, parts o f t h e corpora were used to t est the i n t er-and i n traindividual consistency of the transcribers.The e v aluation of the algorithm currently shows an error rate of 12.87% for read speech a n d 21.03% for spontaneous speech.The i n t erindividual consistency of 95.8% might be considered as an upper limit for any a u t omatic detection method.