Trajectory behaviour at different phonemic context sizes

Abstract

The authors propose a piecewise-linear model for the temporal trajectories of Mel Frequency Cepstral Coefficients during phone transitions. As with conventional Hidden Markov Models, the parameters of the model can be estimated for different phonemic context sizes, but their model allows for an intuitive understanding of the impact of context size. They find that the most detailed models, predictably, match the coefficient tracks best - but when data scarcity forces them to use less detailed models, different types of context modelling (clustered triphones versus biphones) have complimentary behaviours. The authors discuss how this complimentarity may be useful for data-efficient ASR.