Noise-Robust Speech Recognition Based on LPMCC Feature and RBF Neural Network
To solve the problem that recognition rates of speech recognition systems decrease in the noisy environment presently, the Linear Predictive Mel cepstrum coefficient (LPMCC) is used as feature parameter and uses character possessing LPMCC and RBF neural network which have optimal approach capability and the fast training speed, adopts clustering algorithm and entire-supervised algorithm and realizes a noise-robust speech recognition system based on RBF neural net-work. The hidden layer training of clustering algorithm used K-means clustering algorithm and output layer learning used linear least mean square. The adjustment of the entire parameters of entire-supervised algorithm is based on grads decline method. It is a kind of supervised learning algorithm and can choose excellent parameters. Experiments show that entire-supervised algorithm have higher recognition rates in different SNRs than clustering algorithm.
KeywordsSpeech recognition RBF neural network LPCMCC Clustering algorithm Entire-supervised algorithm
- 1.Hou, X., Zhang, X.: A speech recognition method of isolated words based on modified LP cepstrum. J. Taiyuan Univ. Technol. 506–510, 37 (2006)Google Scholar
- 2.Yan, T., Yun, X., Jin, F., Zhu, Q.: RBF neural networks and their application to output–based objective speech quality assessment. Acta Electronica Sinica 1282–1285, 32 (2004)Google Scholar
- 3.Guo, J.J., Luh, P.B.: Selecting input factors for clusters of Gaussian radial basis function network to improve market clearing price prediction. IEEE Trans. Power Syst. 665–672, 18 (2003)Google Scholar
- 4.Jingjiao, L., Jie, S., Li, Z., Tianshun, Y.: Hybrid model of hidden markov models network model in speech recognition. J. Northeast. Univ. (Nat. Sci.), 144–147, 120 (2006). http://www.springer.com/lncs. Accessed 21 Nov 2016
- 6.Hoshimi, M., Niyada, K.: Method and apparatus for speech recognition. J. Acoust. Soc. Am. 109(3), 864 (2018)Google Scholar