Abstract
The paper proposes an approach for noise robust speech recognition in low resource setting. The approach involves formulation of whole word point process model based on word specific spectral peak event in selected groups of mel banks. The performance of the proposed approach is demonstrated on an isolated word recognizer on noisy speech samples (additive white Gaussian noise) at different SNR levels ranging from 0 dB to clean speech. The training is carried out with examples varying from 5 to 80. Performance comparison with HMM based system trained with mel-frequency cepstral coefficients (MFCC) features show an improvement of 8–17% (absolute) depending on SNR level when the number of training examples are less than 10. Since the approach relies only on positions and magnitudes of spectral peaks derived from spoken word examples without any language specific resources, the same can potentially be applied for any language. It is also shown that our approach recognizes those words better that are poorly recognized by HMMs across all SNR levels.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Hinton, G., Deng, L., Yu, D., Dahl, G.E., Mohamed, A.R., Jaitly, N., Senior, A., Vanhoucke, V., Nguyen, P., Sainath, T.N., et al.: Deep neural networks for acoustic modeling in speech recognition: the shared views of four research groups. IEEE Sig. Process. Mag. 29(6), 82–97 (2012)
Jansen, A., Niyogi, P.: Point process models for spotting keywords in continuous speech. IEEE Trans. Audio Speech Lang. Process. 17(8), 1457–1470 (2009)
Jansen, A., Dupoux, E., Goldwater, S., Johnson, M., Khudanpur, S., Church, K., Feldman, N., Hermansky, H., Metze, F., Rose, R., et al.: A summary of the 2012 JHU CLSP workshop on zero resource speech technologies and models of early language acquisition (2013)
Jansen, A., Mesgarani, N., Niyogi, P.: Point process models of spectro-temporal modulation events for speech recognition. In: 2010 Conference Record of the Forty Fourth Asilomar Conference on Signals, Systems and Computers (ASILOMAR), pp. 104–108. IEEE (2010)
Jansen, A., Niyogi, P.: Robust keyword spotting with rapidly adapting point process models. In: Tenth Annual Conference of the International Speech Communication Association (2009)
Jansen, A., Niyogi, P.: Detection-based speech recognition with sparse point process models. In: 2010 IEEE International Conference on Acoustics Speech and Signal Processing (ICASSP), pp. 4362–4365. IEEE (2010)
Liu, C., Jansen, A., Khudanpur, S.: Context-dependent point process models for keyword search and detection-based ASR. In: 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6025–6029. IEEE (2016)
Rabiner, L.R.: A tutorial on hidden Markov models and selected applications in speech recognition. Proc. IEEE 77(2), 257–286 (1989)
Wang, Y., Yang, J.A., Lu, J., Liu, H., Wang, L.W.: Hierarchical deep belief networks based point process model for keywords spotting in continuous speech. Int. J. Commun. Syst. 28(3), 483–496 (2015)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this paper
Cite this paper
Mandal, A., Kumar, K.R.P., Mitra, P. (2017). Point Process Modeling of Spectral Peaks for Low Resource Robust Speech Recognition. In: Ghosh, A., Pal, R., Prasath, R. (eds) Mining Intelligence and Knowledge Exploration. MIKE 2017. Lecture Notes in Computer Science(), vol 10682. Springer, Cham. https://doi.org/10.1007/978-3-319-71928-3_22
Download citation
DOI: https://doi.org/10.1007/978-3-319-71928-3_22
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-71927-6
Online ISBN: 978-3-319-71928-3
eBook Packages: Computer ScienceComputer Science (R0)