Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/273165
Title: | Evaluation of machine learning algorithms based on speech features as predictors to the classification of intellectual disability |
Researcher: | GAURAV AGGARWAL |
Guide(s): | Rekha Vig and Latika Singh |
Keywords: | Engineering and Technology,Computer Science,Computer Science Theory and Methods |
University: | The Northcap University (Formerly ITM University, Gurgaon) |
Completed Date: | Dec 19,2019 |
Abstract: | The present study aims to explore speech as a tool for desining aids that can be used in assisting diagnosis of neurodevelopmental disorders. Speech, which is a fine motor activity, is one of the measureable output of brain. With development of technology, several automated assessments methods are available to extract features of speech. This study aims to use these features to train machine learning algorithms which can differentiate between speech of normal children/adult and children with special needs. Therefore, in this study, several feature extraction techniques, namely, Mel-Frequency Cepstral Coefficients (MFCC), Linear Predictive Cepstral Coefficients (LPCC), Power Spectrum Density, Discrete Cosine Transform (DCT) and Short Time Fourier Transform (STFT) are used to extract the speech features. For each speech sample, a total of 205 features are extracted including 13 acoustical features from MFCC, 128 features from LPCC and 64 features from power spectrum density. Further, Linear predictive coding based parameterization is also applied to each speech sample to extract some more features that are Weighted Linear Predictive Cepstral Coefficients (WLPCC) and Linear Predictive Coding (LPC). For determining the most significant features, feature selection algorithm like Univariate filter approach is also applied to the dataset. newlineA dataset from a government institute SIRTAR and author s institute (The NorthCap University) is created. Classification models such as Support Vector Machine (SVM), Artificial Neural Network (ANN), Radial Basis Function Neural Network (RBFNN, Random Forest, k-Nearest Neighbors (k-NN) and Linear Discriminant Analysis (LDA) are applied to classify the speech samples of children with Intellectual Disability (ID) and Typically Developed (TD) children. Ten-fold cross-validation is used to achieve the reliability of all the classification models. newline |
Pagination: | 119p |
URI: | http://hdl.handle.net/10603/273165 |
Appears in Departments: | Department of CSE & IT |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_title.pdf | Attached File | 191.78 kB | Adobe PDF | View/Open |
02_certificate.pdf | 266.87 kB | Adobe PDF | View/Open | |
03_certificate from the student.pdf | 176.04 kB | Adobe PDF | View/Open | |
04_acknowledgement.pdf | 175.42 kB | Adobe PDF | View/Open | |
05_content.pdf | 304.33 kB | Adobe PDF | View/Open | |
06_figures.pdf | 296.63 kB | Adobe PDF | View/Open | |
07_tables.pdf | 280.98 kB | Adobe PDF | View/Open | |
08_abstract.pdf | 279.87 kB | Adobe PDF | View/Open | |
09_abbreviation.pdf | 180.77 kB | Adobe PDF | View/Open | |
10_chapter 1.pdf | 459.73 kB | Adobe PDF | View/Open | |
11_chapter 2.pdf | 448.57 kB | Adobe PDF | View/Open | |
12_chapter 3.pdf | 6.83 MB | Adobe PDF | View/Open | |
13_chapter 4.pdf | 565.86 kB | Adobe PDF | View/Open | |
14_chapter 5.pdf | 683.09 kB | Adobe PDF | View/Open | |
15_chapter 6.pdf | 187.37 kB | Adobe PDF | View/Open | |
16_references.pdf | 735.5 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: