1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Spoken language identification based on the enhanced self-adjusting extreme learning machine approach
Authors:	Musatafa Abbas Abbood Albadr (P84264)
Supervisor:	Sabrina Tiun, Dr.
Keywords:	Speech processing systems
Issue Date:	14-Jul-2016
Description:	Spoken Language Identification (LID) is the task of determining and classifying natural language from a given content and datasets. Typically, data has to be processed in order to extract useful features for performing LID. Extracting features for LID is mature process in the literature where standard features for LID has been developed starting from Mel-Frequency Cepstral Coefficients (MFCC), Shifted Delta Cepstral (SDC), Gaussian Mixture Model (GMM) and ending with i-vector. However, the process of learning based on the extract features is still yet to be optimized in order to capture all the embedded knowledge in the extracted features. Extreme Learning Machine (ELM) is one effective learning model for performing classification and regression. This model is useful for teaching one hidden layer neural network. However, the learning process of this model is not fully optimized due to the random selection of weights in the input hidden layer. In this study, ELM has been selected as learning model for LID based on standard features extraction. One optimization approach of ELM named as Self-Adjusting Extreme Learning Machine (SA-ELM) has been selected as a benchmark and it has been improved by altering the selection phase of the optimization. The selection has been performed by incorporating both Split-Ratio and K-Tournament, the improved SA-ELM is named Enhanced Self-Adjusting Extreme Learning Machine (ESA-ELM). Results were generated based on LID with eight languages dataset that was also created in this study. The results reveal good superiority in the performance of Enhanced Self-Adjusting Extreme Learning Machine LID (ESA-ELM LID) comparing with SA-ELM LID. The accuracy of ESA-ELM LID was 96.25 %, while the accuracy of SA-ELM LID was 95.00%.,Certification of Master's/Doctoral Thesis" is not available
Pages:	92
Call Number:	TK7882.S65A433 2017 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.