The OpenGrm open-source finite-state grammar software libraries

By Brian Roark, Richard Sproat, Cyril Allauzen, Michael D. Riley, Jeffrey Scott Sorensen, Terry Tai. 2012

120 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Felix Stahlberg
2. Hao Zhang
3. Xu Tan
4. Richard Sproat
5. Courtney Mansfield
6. Kyle Gorman
7. Sevinj Yolchuyeva
8. Navdeep Jaitly

Citations in the corpus

• Algorithms and Theory for Multiple-Source Adaptation (2018, methodology): “We first used the same vocabulary defined for feature extraction to train a separate bigram statistical language model for each domain, using the OpenGrm library [Roark et al., 2012].”
• Encoding linear models as weighted finite-state transducers (2014, methodology): “We used the OpenGrm n-gram library [2] to build joint multi-gram language models encoded as OpenFst [1] format WFSA.”
• Composition-based on-the-fly rescoring for salient n-gram biasing (2015, methodology): “This WFST structure is similar to standard WFST n-gram representations [3], but has some key differences.”
• The SIGMORPHON 2020 Shared Task on Multilingual Grapheme-to-Phoneme Conversion (2020, methodology): “The model is quite similar to the Phonetisaurus toolkit (No-vak et al. 2016), but here is implemented using the OpenGrm toolkit (Roark et al. 2012, Gorman 2016); see Lee et al. 2020 for a full description.”
• Neural Models of Text Normalization for Speech Applications (2019, methodology): “Tokeniza-tion/classification and verbalization grammars are compiled into weighted finite-state transducers (WFSTs) using the Thrax grammar development library (Roark et al. 2012).”
• The Kestrel TTS text normalization system (2014, methodology): “Finite-state grammars for tokenization, classification and verbalization are constructed using an internal toolkit that has also been open-sourced as the Thrax grammar development toolkit; see Tai, Skut and Sproat (2011) and Roark et al. (2012) and http://openfst.cs.nyu.edu/twiki/bin/view/GRM/Thrax.”
• RNN Approaches to Text Normalization: A Challenge (2016, methodology): “We constructed a Thrax grammar (Roark et al., 2012) to cover MEASURE and MONEY expressions, two classes where the RNN is prone to produce silly readings.”
• Neural Text Normalization with Subword Units (2019, methodology): “Traditionally, the task of NSW normalization has been approached by manually authoring grammars in the form of finite-state transducers (Sproat, 1996; Roark et al., 2012) such as integer grammars (e.g., “26” → “twenty six”) or time grammars (e.g., “5:26” → “five twenty six”).”
• Text normalization with convolutional neural networks (2018, methodology): “Weighted finite-state transducers (WFSTs) were also used (Allauzen et al. 2016; Roark et al. 2012).”