Sixth European Conference on Speech Communication and Technology
Landmark based speech processing is a component of Lexical Access From Features (LAFF), a novel paradigm for feature based speech recognition. Detection and classification of landmarks is a crucial first step in a LAFF system. This work implements a Vowel Landmark Detector using a syllabic segmentation algorithm [Mermelstein 75] and examines the relative utility of its several constraints. The detector is scored against the TIMIT database, using a novel algorithm to convert the segmental transcriptions to a landmark representation for scoring. The results show that substantial improvement in performance can be gained by modifying the frequency range for peak detection. An additional advantage of this modification is that post processing to remove fricative peaks is no longer necessary, which substantially simplifies the algorithm.
Full Paper (PDF) Gnu-Zipped Postscript
Bibliographic reference. Howitt, Andrew Wilson (1999): "Vowel landmark detection", In EUROSPEECH'99, 2777-2780.