论文信息 - An Improving MFCC Features Extraction Based on FastICA Algorithm plus RASTA Filtering

An Improving MFCC Features Extraction Based on FastICA Algorithm plus RASTA Filtering

FastICA is a kind of independent component analysis (ICA), which is robust and high performance algorithm, it can strongly remove signal correlation and ensure each signal to be independence. Through perceptual test, improving that RASTA is an idea which can denoise effectively. First, we remove signal correlation through FastICA algorithm, then we use RASTA filter to filtering the ceptral coefficients. Finally, we reduce dimension of the cepstral coefficients by the variances of cepstral coefficients in different dimension and obtain our features. By the HTK3.3, the speech feature extraction which was presented in this paper show the better robust in recognition experiment.

[1] Xihong Wu,et al. On the importance of components of the MFCC in speech and speaker recognition , 2000, INTERSPEECH.

[2] Oh-Wook Kwon,et al. Phoneme recognition using ICA-based feature extraction and transformation , 2004, Signal Process..

[3] Hynek Hermansky,et al. RASTA processing of speech , 1994, IEEE Trans. Speech Audio Process..

[4] Erkki Oja,et al. Independent component analysis: algorithms and applications , 2000, Neural Networks.

[5] Liu Ju,et al. Speech feature extraction based on 2-D independent component analysis , 2007 .

[6] Aapo Hyvärinen,et al. Fast and robust fixed-point algorithms for independent component analysis , 1999, IEEE Trans. Neural Networks.

[7] Wu Shuzhen. A Study for the Application of RASTA on Objective Communication Speech Quality Evaluation , 2003 .

[8] Sarel van Vuuren,et al. Data-driven design of RASTA-like filters , 1997, EUROSPEECH.

[9] M. Mięsikowska. Speech signal processing and analysis tool , 2007 .