Robust Preprocessing and Segmentation of Fluent Speech

Abstract

Emphasis on linguistic distinct features of speech guided research to disjoin prosodic aspects of fluent speech from verbal components. While intonation (together with temporal aspects) is still the prominent carrier of prosody, robust formant tracking remains a key feature of any device providing linguistic analysis of speech (references in Hi91). In conventional approaches (LPC or Cepstrum techniques) where prosodic and non-prosodic aspects of fluent speech are treated separately, prosodic information often conflicts with formant information, as well, as with linguistic distinct temporal information (such as voice onset time or plosive releases).