Grey wolf optimization-extreme learning machine for automatic spoken language identification

Natural language classification and determination based on a particular content and dataset is carried out using Spoken Language Identification (LID) which typically involves the extraction of valuable elements in a mature data processing procedure whereby the regular LID features had been developed using the Mel Frequency Cepstral Coefficient (MFCC), Shifted Delta Coefficient (SDC), Gaussian Mixture Model (GMM) and an i-vector framework. However, there remains a need for optimization in terms of the learning process so as to allow for all the knowledge embedded in the extracted features to be captured completely. A powerful machine learning algorithm known as Extreme Learning Machine (ELM) is used for conducting regression and classification and can train single hidden layer neural networks effectively. Yet, ELM's learning process remains under-optimized owing to the entrenched random weights selection in the input hidden layer. Based on the standard feature extraction, this current study chooses ELM as the learning model for LID. An optimized method known as the Enhanced Self-Adjusting-ELM (ESA-ELM) has been chosen as a benchmark with enhancements via the adoption of an alternate optimization approach i.e., Grey Wolf Optimisation (GWO) rather than Enhanced Ameliorated Teaching Learning-Based Optimization (EATLBO) to ensure higher performance. Ultimately, this enhanced version of the ESA-ELM is referred to as a Grey Wolf Optimisation-Extreme Learning Machine (GWO-ELM). The results generation is carried out based on LID using the exact benchmark dataset that was derived from eight separate languages. The results indicated that the GWO-ELM LID has a much superior performance than the ESA-ELM LID with respective accuracies of 100.00% for the former and merely 96.25% for the latter.