Text-Independent Speaker Verification Using Automatically Labelled Acoustic Segments
| Type of publication: | Inproceedings |
| Citation: | petr98:icslp |
| Booktitle: | International Conference on Spoken Language Processing (ICSLP 98), Sidney, Australia |
| Year: | 1998 |
| Pages: | 536-539 |
| URL: | http://diuf.unifr.ch/people/he... |
| Abstract: | Most of text-independent speaker verification techniques are based on modelling the global probability distribution function (pdf) of speakers in the acoustic vector space. Our paper presents an alternative to this approach with a class-dependent verification system using automatically determined segmental units. Segments are found with temporal decomposition and labelled through unsupervised clustering. The core of the system is based on a set of multi-layer perceptrons (MLP) trained to discriminate between client and an independent set of world speakers. Each MLP is dedicated to work with data segments that were previously selected as belonging to a particular class. The last step of the system is a recombination of MLP scores to take the verification decision. Issues and potential advantages of the segmental approach are presented. Performances of global and segmental approaches are reported on the NIST'98 data (250 female and 250 male speakers), showing promising results for the proposed new segmental approach. Comparison with state of the art system, based on Gaussian Mixture Modelling is also included. |
| Keywords: | Biometrics, MLP, Speaker Verification |
| Authors | |
| Added by: | [] |
| Total mark: | 0 |
|
Attachments
|
|
|
Notes
|
|
|
|
|
|
Topics
|
|
|
|
|