Loading [MathJax]/extensions/MathZoom.js
Support vector machines, Mel-Frequency Cepstral Coefficients and the Discrete Cosine Transform applied on voice based biometric authentication | IEEE Conference Publication | IEEE Xplore

Support vector machines, Mel-Frequency Cepstral Coefficients and the Discrete Cosine Transform applied on voice based biometric authentication


Abstract:

In this paper, the implementation of a Support Vector Machine (SVM) is proposed based on automatic system for voice biometric authentication, recognizing the speaker usin...Show More

Abstract:

In this paper, the implementation of a Support Vector Machine (SVM) is proposed based on automatic system for voice biometric authentication, recognizing the speaker using Mel-Frequency Cepstral Coefficients and the Discrete Cosine Transform. The voice recognition problem can be modeled as a classification problem, where the objective is to obtain the best degree of separability between the classes which represent the voice. Building an automated speech recognition system capable of identifying the speaker, has many techniques using artificial intelligence and general classification at disposal, Support Vector Machines being the one used in this work. The voice samples used are in the Brazilian Portuguese language and had its features extracted through the Discrete Cosine Transform. Extracted features are applied on the Mel-frequency Cepstral Coefficients to create a two-dimensional matrix used as input to the SVM algorithm. This algorithm generates the pattern to be recognized, leading to a reliable speaker identification using few parameters and a small dataset.
Date of Conference: 10-11 November 2015
Date Added to IEEE Xplore: 21 December 2015
ISBN Information:
Conference Location: London, UK

I. Introduction

Support Vector Machine (SVMs) were developed to solve the classification problem, but, recently, have been applied to solve regression ones [2]. The classifiers generated by a Support Vector Machine achieve good results in general, having that capacity of generalization measured by their efficiency on classifying data that does not belong to the training data set. The foundation of Support Vector Machines (SVM) was developed by Vapnik [1] and earned a lot of popularity due its promising characteristics, with better empirical performance. The mathematical formulation uses the Structural Risk Minimization (SRM), that has shown itself superior to the Empirical Risk Minimization (ERM), used by conventional Neural Nets. SRM minimizes an upper limit over the expected risk, while ERM minimizes the error on the training data. This is the difference that leads SVM to have greater generalization capacity, which is the goal of statistical learning.

References

References is not available for this document.