A1 -- LAFF Background

A1 -- LAFF Background

LAFF (Lexical Access From Features) is a system for representing the acoustic cues for features and organizing them for lexical access.

Articulator-free features describe the type of sound produced. Also called Manner features.

LANDMARKS are points in the signal where acoustic cues for these features are most evident.

Three types of Landmark: Vowel, Glide, Consonant

Articulator-bound features describe how articulators were used to produce the sound. Also called Place features, they are attached to Landmarks.

Published work:

  • First presentation: Stevens (Montreal) 1986
  • General overview: Stevens (EuroSpeech Madrid) 1995
  • Database Labeling: Choi et al. (ASA) 1997
  • Consonant Landmark detection: Liu (JASA) 1996
  • Glide Landmark detection: Sun (thesis) 1997
  • Consonant Voicing: Choi (thesis) 1999
  • Place of Articulation for Consonants: Stevens et al. (ICPhS) 1999
  • Nasality: Chen, in progress
Vowel Landmark Detection needs to be done!