The effectiveness of ICA-based representation: Application to speech feature extraction for noise robust speaker recognition
Xin Guo Zou, Peter Jančovič, Ju Liu · 2006
In this paper, we present a mathematical derivation demonstrating that feature representation obtained by us-ing the Independent Component Analysis (ICA) is an ef-fective representation for non-Gaussian signals when being both clean and corrupted by Gaussian noise. Our findings are experimentally demonstrated by employing the ICA for speech feature extraction; specifically, the ICA is used to transform the logarithm filter-bank-energies (instead of the DCT which provides MFCC features). The evaluation is pre-sented for a GMM-based speaker identification task on the TIMIT database for clean speech and speech corrupted by white noise. The effectiveness of ICA is analysed individu-ally for signals corresponding to each phoneme. The experi-mental results show that the ICA-based features can provide significantly better performance than traditional MFCCs and PCA-based features in both clean and noisy speech. 1.