Bayesian joint-sequence models for grapheme-to-phoneme conversion

Bayesian joint-sequence models for grapheme-to-phoneme conversion

Paper in proceedings (conference paper)

We describe a fully Bayesian approach to grapheme-to-phonemeconversion based on the joint-sequence model (JSM). Usually, standardsmoothed n-gram language models (LM, e.g. Kneser-Ney)are used with JSMs to model graphone sequences (joint graphemephonemepairs). However, we take a Bayesian approach using ahierarchical Pitman-Yor-Process LM. This provides an elegant alternativeto using smoothing techniques to avoid over-training. Noheld-out sets and complex parameter tuning is necessary, and severalconvergence problems encountered in the discounted Expectation-Maximization (as used in the smoothed JSMs) are avoided. Everystep is modeled by weighted finite state transducers and implementedwith standard operations from the OpenFST toolkit. Weevaluate our model on a standard data set (CMUdict), where it givescomparable results to the previously reported smoothed JSMs interms of phoneme-error rate while requiring a much smaller training/testing time. Most importantly, our model can be used in aBayesian framework and for (partly) un-supervised training.

We describe a fully Bayesian approach to grapheme-to-phonemeconversion based on the joint-sequence model (JSM). Usually, standardsmoothed n-gram language models (LM, e.g. Kneser-Ney)are used with JSMs to model graphone sequences (joint graphemephonemepairs). However, we take a Bayesian approach using ahierarchical Pitman-Yor-Process LM. This provides an elegant alternativeto using smoothing techniques to avoid over-training. Noheld-out sets and complex parameter tuning is necessary, and severalconvergence problems encountered in the discounted Expectation-Maximization (as used in the smoothed JSMs) are avoided. Everystep is modeled by weighted finite state transducers and implementedwith standard operations from the OpenFST toolkit. Weevaluate our model on a standard data set (CMUdict), where it givescomparable results to the previously reported smoothed JSMs interms of phoneme-error rate while requiring a much smaller training/testing time. Most importantly, our model can be used in aBayesian framework and for (partly) un-supervised training.

Bayesian approach, joint-sequence models,weighted finite state transducers, letter-to-sound, grapheme-tophoneme conversion, hierarchical Pitman-Yor-Process

Bayesian approach, joint-sequence models,weighted finite state transducers, letter-to-sound, grapheme-tophoneme conversion, hierarchical Pitman-Yor-Process

HANNEMANN, M.; TRMAL, J.; ONDEL YANG, L.; KESIRAJU, S.; BURGET, L.

2018

05.03.2017

IEEE Signal Processing Society

New Orleans

978-1-5090-4117-6

Proceedings of ICASSP 2017

2836

2840

5

https://www.fit.vut.cz/research/publication/11469/

hannemann_icassp2017_0002836