Comparative Study of Speech Recognition System Using Various Feature Extraction Techniques

It is very important to detect the speech endpoints accurately in speech recognition. This paper presents a comparative analysis of various feature extraction techniques of endpoint detection in speech recognition of isolated words in noisy environments. The endpoint detection problem is nontrivial for no stationary backgrounds where artifacts (i.e., no speech events) may be introduced by the speaker, the recording environment, and the transmission system. An optimum set of characteristics is identified by combining parameters from both time domain and frequency domain, in a robust approach for identification when the speech signal is corrupted by additive noise and channel distortion. The cases of colored noises such as babble noise, factory noise at different SNR values in conjunction with distortions due to recording medium were tested. Experimental results identify the optimal algorithm which significantly achieves the highest performance in the recognition task.