论文信息 - Using Phoneme Segmentation in Conjunction with Missing Feature Approaches for Noise Robust Speech Recognition

Using Phoneme Segmentation in Conjunction with Missing Feature Approaches for Noise Robust Speech Recognition

Cluster-based reconstruction is a feature based method that shown promising results in improvement of speech recognition accuracy but in low SNR values and multiple clusters, classification of noisy vectors is badly degrade the recognition accuracy. Main idea of this paper is to take advantage of phonetic properties and phonetic clustering to overcome disadvantage of classification step. We proposed three different clustering strategies in order to solve clustering misclassification problem and improve speech recognition accuracy in presence of additive noise through Phoneme Segmentation in conjunction with Missing Feature approaches. Third method results show an average improvement of 14.4% in 0 dB and 8.35% in -10 dB in comparison with conventional cluster-based reconstruction.

A. Mohammadi | F. Almasganj | F. Naderkhani | A. Taherkhani

[1] Richard M. Stern,et al. Reconstruction of incomplete spectrograms for robust speech recognition , 2000 .

[2] Phil D. Green,et al. ROBUST ASR WITH UNRELIABLE DATA AND MINIMAL ASSUMPTIONS , 1999 .

[3] Michael K. Jensen,et al. Speech Reconstruction from Binary Masked Spectrograms Using Vector Quantized Speaker Models , 2006 .

[4] Richard M. Stern,et al. Reconstruction of missing features for robust speech recognition , 2004, Speech Commun..

[5] D. Rubin,et al. Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper , 1977 .

[6] Lie Lu,et al. Speech segmentation without speech recognition , 2003, 2003 International Conference on Multimedia and Expo. ICME '03. Proceedings (Cat. No.03TH8698).