Language Model Optimization for a Deep Neural Network Based Speech Recognition System for Serbian

  • Edvin PakociEmail author
  • Branislav Popović
  • Darko Pekar
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 10458)


This paper presents the results obtained using several variants of trigram language models in a large vocabulary continuous speech recognition (LVCSR) system for the Serbian language, based on the deep neural network (DNN) framework implemented within the Kaldi speech recognition toolkit. This training approach allows parallelization using several threads on either multiple GPUs or multiple CPUs, and provides a natural-gradient modification to the stochastic gradient descent (SGD) optimization method. Acoustic models are trained over a fixed number of training epochs with parameter averaging in the end. This paper discusses recognition using different language models trained with Kneser-Ney or Good-Turing smoothing methods, as well as several pruning parameter values. The results on a test set containing more than 120000 words and different utterance types are explored and compared to the referent results with GMM-HMM speaker-adapted models for the same speech database. Online and offline recognition results are compared to each other as well. Finally, the effect of additional discriminative training using a language model prior to the DNN stage is explored.


Deep neural networks Kaldi Serbian Language modeling MMI 



. The work described in this paper was supported in part by the Ministry of Education, Science and Technological Development of the Republic of Serbia, within the project “Development of Dialogue Systems for Serbian and Other South Slavic Languages”, EUREKA project DANSPLAT, “A Platform for the Applications of Speech Technologies on Smartphones for the Languages of the Danube Region”, ID E! 9944, and the Provincial Secretariat for Higher Education and Scientific Research, within the project “Central Audio-Library of the University of Novi Sad”, No. 114-451-2570/2016-02.


  1. 1.
    Popović, B., Pakoci, E., Ostrogonac, S., Pekar, D.: Large vocabulary continuous speech recognition for Serbian using the Kaldi toolkit. In: Proceedings of the 10th Conference on Digital Speech and Image Processing, DOGS, Novi Sad, pp. 31–34 (2014)Google Scholar
  2. 2.
    Popović, B., Ostrogonac, S., Pakoci, E., Jakovljević, N., Delić, V.: Deep neural network based continuous speech recognition for Serbian using the Kaldi toolkit. In: Ronzhin, A., Potapova, R., Fakotakis, N. (eds.) SPECOM 2015, LNCS, vol. 9319, pp. 186–192. Springer, Cham (2015)Google Scholar
  3. 3.
    Povey, D., Kuo, H.-K.J., Soltau, H.: Fast speaker adaptive training for speech recognition. In: Proceedings of the 9th Annual Conference of the International Speech Communication Association, INTERSPEECH, Brisbane, pp. 1245–1248 (2008)Google Scholar
  4. 4.
    Povey, D.: Discriminative training for large vocabulary speech recognition. Ph.D. thesis, University, Engineering Department, Cambridge (2003)Google Scholar
  5. 5.
    Povey, D., Woodland, P.C.: Minimum phone error and I-smoothing for improved discriminative training. In: Proceedings of the 27th International Conference on Acoustics, Speech and Signal Processing, ICASSP, Orlando, pp. I-105–I-108 (2002)Google Scholar
  6. 6.
    Suzić, S., Ostrogonac, S., Pakoci, E., Bojanić, M.: Building a speech repository for a Serbian LVCSR system. Telfor J. 6(2), 109–114 (2014). Paunović, Đ., Milić, L. (Eds.) Telecommunications Society, BelgradeCrossRefGoogle Scholar
  7. 7.
    Stolcke, A., Zheng, J., Wang, W., Abrash, V.: SRILM at sixteen: update and outlook. In: IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU, Waikoloa, p. 5 (2011)Google Scholar
  8. 8.
    Kneser, R., Ney, H.: Improved backing-off for M-gram language modeling. In: Proceedings of the 20th International Conference on Acoustics, Speech and Signal Processing, ICASSP, Detroit, pp. 181–184 (1995)Google Scholar
  9. 9.
    Gale, W.A., Sampson, G.: Good-Turing smoothing without tears. J. Quant. Linguist. 2(3), 217–237 (1995). Köhler, R. (Ed.) Swets & Zeitlinger, LisseCrossRefGoogle Scholar
  10. 10.
    Povey, D., Kanevsky, D., Kingsbury, B., Ramabhadran, B., Saon, G., Visweswariah, K.: Boosted MMI for model and feature-space discriminative training. In: Proceedings of the 33rd International Conference on Acoustics, Speech and Signal Processing, ICASSP, Las Vegas, pp. 4057–4060 (2008)Google Scholar
  11. 11.
    Povey, D., Zhang, X., Khudanpur, S.: Parallel training of DNNs with natural gradient and parameter averaging. In: Proceedings of the 3rd International Conference on Learning Representations Workshop, ICLR, San Diego (2015)Google Scholar
  12. 12.
    Peddinti, V., Povey, D., Khudanpur, S.: A time delay neural network architecture for efficient modeling of long temporal contexts. In: Proceedings of the 14th Annual Conference of the International Speech Communication Association, INTERSPEECH, Dresden, pp. 2–6 (2015)Google Scholar
  13. 13.
    Bhanuprasad, K., Svenson, D.: Errgrams – a way to improving ASR for highly inflective Dravidian languages. In: Proceedings of the 3rd International Joint Conference on Natural Language Processing, IJCNLP, Hyderabad, pp. 805–810 (2008)Google Scholar

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  1. 1.Faculty of Technical Sciences, Department for Power, Electronic and Telecommunication EngineeringUniversity of Novi SadNovi SadSerbia
  2. 2.AlfaNum Speech TechnologiesNovi SadSerbia

Personalised recommendations