An investigation of acoustic events related to sentential stress and pitch accents, in English

Abstract An algorithm is described which abstracts acoustic parameters of a speech waveform to automatically transcribe sentential stress and pitch movements. The waveform acoustics used are duration, energy and fundamental frequency. The abstractions described aim to isolate the prosodically imposed variations in these parameters. A method of syllabification from acoustic parameters is presented. The prominence of each syllable is determined using the automatic process described and the resultant transcription is compared with a hand-labelled prosodic transcription. The agreement level of 61.6% suggests that acoustic parameters other than those already used by the algorithm may be available to the human labeller.