An Evaluation of MFCC and ZCR Feature Extraction with CNN Model of Speaker Identification System

The speech is the important medium to transfer our feeling to the neighbors. Speaker recognition is the process to recognize our speech in our owned database. Speaker recognition has divided in to speaker identification and speaker verification. Speaker identification is different mode which is text dependent, independent and prompted. Identifying the speaker in our database is the major task in our process. Identifying data or speech is the main aim to handle in ourdatabase. Speaker identification skills arebroadly useful in security and investigation, voice confirmation, electronic voice snooping and classify verification. Various features extraction ideas are used in identification as well as verification such as mfcc, zcr, pncc and so on. MFCC are cepstral coefficients resultant on a warped frequency scale centered on human hearing observation. In the computation of MFCC, the first thing is windowing the speech signal to split the speech signal into frames. The zero-crossing rate is the proportion at which the signal modifications from positive to zero to negative or from negative to zero to positive. Power Normalized Cepstral Coefficients is well identified for the high accuracy of automatic speech recognition systems even in high-noise situations .This paper presents the comparison of mfcc and zcr with the CNN modeling system. CNN is the top modeling idea in the recent decades.