Development of Assamese Continuous Speech Recognition System

Barsha Deka, Nirmala S.R., Samudravijaya K.


´╗┐This paper describes the development of a continuous speech recognition system for Assamese, an under-resourced language of North-East India. The Speech corpus used in this work consists of 5658 spoken utterances collected from 27 speakers over telephone channel. The baseline speech recognition system was implemented using conventional hidden Markov model in conjunction with Gaussian mixture model, employing Mel-frequency cepstral coefficients as features. ASR systems using subspace Gaussian mixture model and deep neural networks together with hidden Markov model were implemented. The systems were evaluated with 3-fold cross validation method. The average word error rate of the best ASR system is 4.3%.


 DOI: 10.21437/SLTU.2018-46

Cite as: Deka, B., S.R., N., K., S. (2018) Development of Assamese Continuous Speech Recognition System. Proc. The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages, 220-224, DOI: 10.21437/SLTU.2018-46.


@inproceedings{Deka2018,
  author={Barsha Deka and Nirmala S.R. and Samudravijaya K.},
  title={{Development of Assamese Continuous Speech Recognition System}},
  year=2018,
  booktitle={Proc. The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages},
  pages={220--224},
  doi={10.21437/SLTU.2018-46},
  url={http://dx.doi.org/10.21437/SLTU.2018-46}
}