Constraint-Based Acoustic Modelling

This book is concerned with the automatic generation of inheritance hierarchies for the level below the phoneme and their application in acoustic modelling for automatic speech recognition. Over the last decade, computational phonology has adopted the typed feature structure representation formalism while paying little attention to subphonetic feature structures. Similarly, paradigmatic phonological knowledge used in acoustic modelling does usually not go beyond phonological feature bundles. Here, the discrepancy between hierarchically-structured lexica, as they are promoted in computational phonology and the flat-structured feature bundles common in acoustic modelling, is addressed in a powerful framework for solving combinatorial problems, i.e. constraint programming.