An investigation into language model data augmentation for low-resourced STT and KWS

Guangpu Huang (1) , Thiago Fraga da Silva (1) , Lori Lamel (1) , Jean-Luc Gauvain (1) , Arseniy Gorin (1) , Antoine Laurent (1) , Rasa Lileikyté (1) , Abdelkhalek Messaoudi (1)

1 LIMSI - Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur

This paper reports on investigations using two techniques for language model text data augmentation for low-resourced automatic speech recognition and keyword search. Low- resourced languages are characterized by limited training materials, which typically results in high out-of-vocabulary (OOV) rates and poor language model estimates. One technique makes use of recurrent neural networks (RNNs) using word or subword units. Word-based RNNs keep the same system vocabulary, so they cannot reduce the OOV, whereas subword units can reduce the OOV but generate many false combinations. A complementary technique is based on automatic machine translation, which requires parallel texts and is able to add words to the vocabulary. These methods were assessed on 10 languages in the context of the Babel program and NIST OpenKWS evaluation. Although improvements vary across languages with both methods, small gains were generally observed in terms of word error rate reduction and improved keyword search performance.

Mots clés: multilingual; low resourced languages; speech recognition; keyword search

Domaines: Informatique [cs]; Informatique et langage [cs.CL]

Guangpu Huang, Thiago Fraga da Silva, Lori Lamel, Jean-Luc Gauvain, Arseniy Gorin, et al.. An investigation into language model data augmentation for low-resourced STT and KWS. IEEE International Conference on Acoustics, Speech, and Signal Processing, IEEE, Mar 2017, New Orleans, United States. ⟨hal-01837171⟩