A module for acoustic-phonetic transcription of fluently spoken German speech

This system is a module of a speech understanding system developed at our institute. After a low-pass filtering and sampling, the speech signal is divided into overlapping frames. These frames are classified into phonetic classes. With a set of rules, adjacent frames corresponding to a phone are combined into segments. The output is a string of segments containing several alternative phones, together with their matching score and prosodic features. An example and results are presented.