A default-and-refinement approach to pronunciation prediction
MH Davel, Etienne Barnard · 2004
We define a novel g-to-p prediction algorithm that utilises the concept of a ‘default phoneme’: a grapheme which is realised as a specific phoneme significantly more often than as any other phoneme. We find that this approach results in an algorithm that performs well across a range from very small to large data sets. We evaluate the algorithm on two benchmarked databases (Fonilex and NETtalk) and find highly competitive performance in asymptotic accuracy, initial learning speed, and model compactness.