Speaker Recognition and Performance Comparison based on Machine Learning
Rajeev Ranjan · Turkish Journal of Computer and Mathematics Education (TURCOMAT) · 2021
The speaker recognition is one of the most emerging field in the area of digital signalprocessing. In this paper, analyzed the performance comparison of voice recognition using Mel frequencyCepstrum coefficient (MFCC) with two methods based on vector quantization (VQ) techniques by Lindebuzo gray (LBG) algorithm and improved weighted VQ method. In the first phase, extract the minoramount of data points from the voice signal that is used subsequent to represent every speaker is known asMFCC and in the second phase, for feature matching there are two approaches which have been proposedhere based on VQ techniques for recognition purpose and also comparison of both algorithms are beingdone with different time length speech samples to get better recognition rate and improving in efficiency ofthe system. The VQ is used as feature matching with traditional LBG Algorithm and improved weightedVQ algorithm for better recognition. The weight of vector in VQ algorithm to get the weighted distortionand after compare between traditional and improved weighted VQ algorithm. When the test time ‘1s’ fortraditional VQ algorithm 81.25% whereas improved weighted VQ 87.5 % and if test time ‘2s’ fortraditional VQ algorithm 87 % whereas, improved weighted VQ 93.7 %. That is improved weighted VQalgorithm gives the better speaker recognition than VQ with LBG algorithm.