WFST-Based Grapheme-to-Phoneme Conversion: Open Source tools for Alignment, Model-Building and Decoding
Josef R. Novak, Nobuaki Minematsu, Keikichi Hirose · 2012
This paper introduces a new open source, WFST-based toolkit for Grapheme-to-Phoneme conversion. The toolkit is efficient, accurate and currently supports a range of features including EM sequence alignment and several decoding techniques novel in the context of G2P. Experimental results show that a combination RNNLM system outperforms all previous reported results on several standard G2P test sets. Preliminary experiments applying Lattice Minimum Bayes-Risk decoding to G2P conversion are also provided. The toolkit is implemented using OpenFst.