Detection of focal points in speech prosody

Signal processing tools are developed to help automatically detect significant events or focal points in the speech and gesture traces of audio-visual data and investigate their temporal correlation as part of a multi-disciplinary effort to assemble the computational resources for facilitating research in gesture and speech interaction. One key task in this research is to determine focal points in the speech signal by analyzing its acoustic-prosodic features. This task requires processing and analysis of an immense amount of data. In order to make the task efficient and less time consuming, algorithms for speech processing are developed to perform automatic feature extraction, parametric representation, and detection of focal points that are usually painstakingly marked by hand. In this paper a method for detecting the focal points in speech prosody is presented and the results are compared with focal points hand-marked by experts. The method produces a high rate of successful detection.