Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Multi-disciplinary Studies | en_US |
| dc.contributor | Department of Electronic Engineering | en_US |
| dc.creator | Yuen, Chi-leung | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/4624 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | - |
| dc.rights | All rights reserved | en_US |
| dc.title | RASTA channel compensation for speaker verification system | en_US |
| dcterms.abstract | The environmental factors tend to degrade the performance of speaker verification systems. The degree of degradation varies with the speech analysis methods, the channel compensation methods and the verification techniques. By using the YOHO corpus [4] and Elliptical Basis Function (EBF) neural networks [1] [2] [3], the robustness and reliability of three different feature extraction methods were evaluated. The features include (a) Perceptual Linear Predictive (PLP) coefficients [6], (b) LP-derived Cepstrum (CEP) coefficients [7], and (c) Relative Spectral-Perceptual Linear Predictive (RASTA-PLP) coefficients [5]. Experimental results show that the performance of the PLP method is superior to the other two methods. For example, the PLP method achieves a verification error rate of 14.9%, whereas the error rates are 22.2% and 27.5% for methods using the RASTA-PLP coefficients and the CEP coefficients, respectively. Results also demonstrate that optimal EBF network parameters for clean speech does not necessarily lead to optimal performance for channel distorted speech. For different prediction orders, a separated set of network parameters is essential for achieving optimal verification performance. Lastly, this study finds that the best Equal Error Rate (EER) does not bring about the best verification performance for a speaker recognition system. However, the EER can be used to measure the degree of overlapping among feature clusters and to compare the decision boundaries' effectiveness in separating speaker features. | en_US |
| dcterms.extent | 83 leaves : ill. ; 31 cm | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 1999 | en_US |
| dcterms.educationalLevel | All Master | en_US |
| dcterms.educationalLevel | M.Sc. | en_US |
| dcterms.LCSH | Speech processing systems, | en_US |
| dcterms.LCSH | Automatic speech recognition | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | restricted access | en_US |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| b14854053.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 2.58 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/4624