Speech Enhancement for Speaker Identification
R. Mahesh · 2018
This paper presents a algorithm for speech enhancement and speaker identification. Discrete-wavelet transform (DWT) based algorithm are used for speech enhancement. In this a Five level discrete wavelet decomposition is used and all the five level output are given as input to the binary mask thresholding function to enhance the speech patterns. The speaker identification process is based on support vector machine. MFCC feature is extracted from the enhanced speech and is used as the speaker features. Output Signal to Noise Ratio and cepstrum distance of the enhanced speech signal is calculated & compared using different wavelet transform. The results obtained are based on processing a NOIZEUS database of reference speech signals of different noises such as train noise, car noise, station noise in several SNRs.