US 12,230,245 B2
Text normalization and inverse text normalization using weighted finite-state transducers and neural language models
Evelina Bakhturina, Santa Clara, CA (US); Yang Zhang, New York, NY (US); and Boris Ginsburg, Sunnyvale, CA (US)
Assigned to Nvidia Corporation, Santa Clara, CA (US)
Filed by Nvidia Corporation, Santa Clara, CA (US)
Filed on Aug. 31, 2022, as Appl. No. 17/900,310.
Prior Publication US 2024/0071366 A1, Feb. 29, 2024
Int. Cl. G06F 17/00 (2019.01); G06F 40/40 (2020.01); G10L 13/047 (2013.01); G10L 13/08 (2013.01)
CPC G10L 13/08 (2013.01) [G06F 40/40 (2020.01); G10L 13/047 (2013.01); G10L 2013/083 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
receiving an input including one or more tokens;
generating, using one or more rule-based algorithms, a set of one or more plain text representations corresponding to the one or more tokens;
determining, using the one or more rule-based algorithms and for at least one individual plain text representation of the set of one or more plain text representations, one or more weights;
determining, from the set of one or more plain text representations and based at least in part on comparing the one or more weights to one or more thresholds, a subset of one or more plain text representations; and
selecting, using a trained language model, a plain text representation from the subset of one or more plain text representations.