雲间烟火Efficient Acoustic Front-End Processing for Tamil Speech Recognition using Modified GFCC Features
were implemented. The results proved that the GFCC features are best suited for all the recognition techniques involved in this work. Further, the best methods have been chosen and the proposed techniques were applied and its performances are verified.
One pre-processing technique and three different feature extraction technique were proposed namely MTYW-GFCC, MTYW-GFCC-FF and FWCMN-MTYW-GFCC-FF. The proposed pre-processing and feature extraction techniques has moderately increased the performance of the Tamil speech recognition based on WRR and RTF. Better results were achieved with HMM, MLP and SVM techniques with increased accuracy. It is also observed that the proposed methods have reduced the testing time for MLP and SVM techniques. Based on the improvements achieved with the proposed methods, this research work will be further extended for Tamil ASR under different noisy conditions in future.
