Hybrid Feature Selection Method Based on Harmony Search and Naked Mole-Rat Algorithms for Spoken Language Identification From Audio Signals

Abstract

This era is dominated by artificial intelligence and its various applications-one of which is Spoken Language Identification (S-LID) which has always been a challenging issue and an important research area in the domain of speech signal processing. This paper deals with SLID to be used for Human-Computer Interaction (HCI) based applications by attempting to classify various languages from three multilingual databases namely CSS10: A Collection of Single Speaker Speech Datasets for 10 Languages, VoxForge and Indian Institute of Technology, Madras (IIT-Madras) speech corpus database by extracting their Mel-Spectrogram features and Relative Spectral Transform-Perceptual Linear Prediction (RASTA-PLP) features. A new hybrid Feature Selection (FS) algorithm have been developed using the versatile Harmony Search (HS) algorithm and a new nature-inspired algorithm called Naked Mole-Rat (NMR) algorithm to select the best subset of features and reduce the model complexity to help it train faster. This selected feature set is fed to five classifiers namely Support Vector Machine (SVM), k-Nearest Neighbor (k-NN), Multi-layer Perceptron (MLP), Naïve Bayes (NB) and Random Forest (RF). The evaluation measures used in this paper are precision, recall, f1-score, classification accuracy and number of selected features. An accuracy of 99.89% on CSS10, 98.22% on VoxForge and 99.75% on IIT-Madras speech corpus databases is achieved using RF. Furthermore, the proposed algorithm is found to outperform 15 standard meta-heuristic FS algorithms.

A Tri-Stage Wrapper-Filter Feature Selection Framework for Disease Classification

“…To outplay this issue, researchers have been working on different kinds of feature selection methods in order to obtain the optimal combination of features. Besides feature selection of medical datasets [ 32 , 33 , 34 ], researchers have utilized feature selection techniques in various domains such as handwritten script classification [ 35 , 36 ], facial emotion recognition [ 37 ], speech emotion recognition [ 38 ], and spoken language identification from audio signals [ 39 , 40 ] and have achieved notable classification accuracy improvement over the years. However, the two-phase filtering with the combination of four kinds of filter methods and classification, as well as one more phase of wrapper algorithm for the mentioned datasets have not been explored thus far.…”

Improving Indian Spoken-Language Identification by Feature Selection in Duration Mismatch Framework

“…Over the years, several language-independent acoustic features like Shifted Delta Cepstral coefficients (SDC) [2,3], Mel Frequency Cepstral Coefficients (MFCC) [4,5], Linear Predictive Coefficients (LPC) [3], Perceptual Linear Prediction (PLP) [6] are reported to perform better for same train-test duration utterances. Although probabilistic linear discriminant analysis (PLDA) based i-vector with modified prior estimation technique [7] and exemplarbased technique [8] was reported to improve the performance of SLID system in duration mismatched conditions, it was not significant, especially for short duration utterances [9,10].…”

“…An optimum feature set of 972 and 1141 selected for IITM and IIT-H data sets reported accuracies 92.35% and 100% with computation time of 158 and 182 min, respectively. Guha et al [6] [3,16].…”

“…The state-of-the-art SLID systems used Vector Quantization (VQ), Gaussian Mixture Model (GMM) [4,17], Support Vector Machine (SVM) [18][19][20], Hidden Markov Model (HMM) [21], Artificial Neural Network (ANN) [4,21,22], and Random Forest (RF) [3,6]. The modern endto-end language recognition models based on deep learning (DL) algorithms improves the performance by increasing the data set requirement and do not perform well for small data set.…”

“…It can be observed that each language cluster is more distinct for duration normalized features than original openS-MILE features, indicating an improved feature subset for classification. [1], [2], [3], [4], [5], [7] linregerrA, linregerrQ, linregerrQ, pctlrange0-1, percentile1.0, stddev 20 lspFreq-sma[0], [1], [4], [6] amean, percentile1.0, 99.0, quartile1,2,3, skewness, upleveltime75 3 lspFreq-sma-de [6] iqr2-3,linregerrA, quartile1, 35 mfcc-sma[0], [1], [2], [3], [5], [6], [7], [8] amean, linregc2, pctlrange0-1, percentile1.0,99.0, quartile1,2,3 7 mfcc-sma-de[0], [1], [6] percentile1.0, linregerrA, linregerrQ, stddev 1 pcm-loudness-sma amean 10 pcm-loudness-sma-de iqr1-2, iqr1-3, iqr2-3, linregerrA, linregerrQ, pctlrange0-1, percentile1.0, 99.0, quartile1,3, stddev 10 pcm-loudness-sma iqr1-3, linregc2, linregerrA, linregerrQ, pctlrange0-1, percentile1.0, 99.0, quartile1,2,3, stddev 2 shimmerLocal-sma quartile1,2…”