Noise Robust Tamil Speech Word Recognition System by Means of PAC Features with ANFIS

In the prior (earlier) speech word recognition system, the speech words are recognized from the input speech words using ANFIS. But this method performance has to be improved in terms of their accuracy and noise robust of the speech recognition. To improve the performance a new Tamil speech word recognition system is proposed with Phase Autocorrelation (PAC). In our proposed system, PAC features are extracted from the input speech word signals. In PAC the features are extracted from the PAC spectrum are called PAC features. The extracted features from the PAC spectrum are Energy entropy, Zero crossing rate and short time energy. Afterward, the extracted PAC features from the feature extraction phase are given to the recognition. In recognition, an ANFIS system is utilized to check whether the input Tamil speech words are recognized or unrecognized. In word recognition, the ANFIS system is well trained by the features from feature extraction process and the recognition performance is validated by utilizing a set of testing speech words. The implementation and the comparison result shows that our proposed system has given high recognition rate in different noise levels.