International Journal For Multidisciplinary Research

E-ISSN: 2582-2160   •   Impact Factor: 9.24

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Call for Paper Volume 8, Issue 6 (November-December 2026) Submit your research before last 3 days of December to publish your research paper in the issue of November-December.

Comparative Analysis of Discrete Wavelet Transforms and Deep Sequence Classifiers for Robust Text-Independent Speaker Identification

Author(s) Mr. Ajaykumar Patil
Country India
Abstract Speaker independent automated Speaker Identification (ASI) systems are dependent on techniques of acoustic feature extraction that preserve the speaker perception of some vocal tract characteristics of the speaker while minimising the effects of background noise, inter-session variation and phonetic differences [1] [2]. Mel-Frequency Cepstral Coefficients (MFCCs), which are the most commonly used traditional acoustic features obtained from the Short-Time Fourier Transform (STFT), have been found to be uniformly time-frequency windowed and the information lost in the higher frequencies [4, 11]. To overcome these dynamic limits, this research aims to make an empirical study about Multiresolution Discrete Wavelet Transforms (DWT) combined with statistical machine learning and deep sequential classification models for text independent speaker identification [5] and [9]. The three different families of mother wavelets are Daubechies (db4), Symlets (sym5), and Coiflets (coif3) with a decomposition structure of 5-level dyadic decomposition (D_{1}–D_{5} and A_{5}) [10, 12]. A statistical vector of 24 dimensions (logarithmic energy, variance, Shannon entropy and kurtosis) is derived from each frame's decomposed sub-bands [12], [13]. The classification abilities of these wavelet signatures are systematically compared by using Support Vector Machines with a Radial Basis Function kernel (SVM-RBF) [15], 1-D Convolutional Neural Networks (1D-CNN) [5, 14] and Bidirectional Long Short-Term Memory networks (Bi-LSTM) [4, 16]. The Daubechies (db4) wavelet paired with an SVM-RBF classifier performed very well in terms of identification accuracy (83.33%, compared to 73.46% for deep sequence models under the same conditions [6, 15]).
Keywords Automatic Speaker Identification, Discrete Wavelet Transform, Multiresolution Analysis, Support Vector Machine, 1-D Convolutional Neural Network, Bidirectional LSTM, LibriSpeech,
Field Engineering
Published In Volume 4, Issue 6, November-December 2022
Published On 2022-11-10
DOI https://doi.org/10.36948/ijfmr.2022.v04i06.91025

Share this