ISCA Archive Interspeech 2013
ISCA Archive Interspeech 2013

Discriminatively trained dependency language modeling for conversational speech recognition

Benjamin Lambert, Bhiksha Raj, Rita Singh

We present a discriminatively trained dependency parser-based language model. The model operates on utterances, rather than words, and so can utilize long-distance structural features of each sentence. We train the model discriminatively on n-best lists, using the perceptron algorithm to tune the model weights. Our features include standard n-gram style features, long-distance co-occurrence features, and syntactic structural features. We evaluate this model by re-ranking n-best lists of recognized speech from the Fisher dataset of informal telephone conversations. We compare various combinations of feature types, and methods of training the model.


doi: 10.21437/Interspeech.2013-748

Cite as: Lambert, B., Raj, B., Singh, R. (2013) Discriminatively trained dependency language modeling for conversational speech recognition. Proc. Interspeech 2013, 3414-3418, doi: 10.21437/Interspeech.2013-748

@inproceedings{lambert13_interspeech,
  author={Benjamin Lambert and Bhiksha Raj and Rita Singh},
  title={{Discriminatively trained dependency language modeling for conversational speech recognition}},
  year=2013,
  booktitle={Proc. Interspeech 2013},
  pages={3414--3418},
  doi={10.21437/Interspeech.2013-748}
}