A Relevance Score Estimation for Spoken Term Detection Based on RNN-Generated Pronunciation Embeddings

Jan Švec, Josef V. Psutka, Luboš Šmídl, Jan Trmal


In this paper, we present a novel method for term score estimation. The method is primarily designed for scoring the out-of-vocabulary terms, however it could also estimate scores for in-vocabulary results. The term score is computed as a cosine distance of two pronunciation embeddings. The first one is generated from the grapheme representation of the searched term, while the second one is computed from the recognized phoneme confusion network. The embeddings are generated by specifically trained recurrent neural network built on the idea of Siamese neural networks. The RNN is trained from recognition results on word- and phone-level in an unsupervised fashion without need of any hand-labeled data. The method is evaluated on the MALACH data in two languages, English and Czech. The results are compared with two baseline methods for OOV term detection.


 DOI: 10.21437/Interspeech.2017-1087

Cite as: Švec, J., Psutka, J.V., Šmídl, L., Trmal, J. (2017) A Relevance Score Estimation for Spoken Term Detection Based on RNN-Generated Pronunciation Embeddings. Proc. Interspeech 2017, 2934-2938, DOI: 10.21437/Interspeech.2017-1087.


@inproceedings{Švec2017,
  author={Jan Švec and Josef V. Psutka and Luboš Šmídl and Jan Trmal},
  title={A Relevance Score Estimation for Spoken Term Detection Based on RNN-Generated Pronunciation Embeddings},
  year=2017,
  booktitle={Proc. Interspeech 2017},
  pages={2934--2938},
  doi={10.21437/Interspeech.2017-1087},
  url={http://dx.doi.org/10.21437/Interspeech.2017-1087}
}