Publication: Influencia de los segmentos del discurso en la discriminación del locutor
Loading...
Identifiers
Publication date
2013
Defense date
2014-02-06
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
La autenticación de la identidad de las personas es hoy en día una tarea
crucial, ya que una amplia variedad de sistemas precisan de un método fiable, bien
para determinar o bien para confirmar la identidad de los individuos.
Entre los métodos de autenticación, el “reconocimiento biométrico” ha
recibido una considerable atención en los últimos años debido principalmente a
dos motivos: el fuerte crecimiento de la demanda de aplicaciones de seguridad,
tanto comerciales como militares y el rápido desarrollo de la tecnología que las
soporta. Su finalidad es la determinación de la identidad de las personas
basándose en uno o más rasgos físicos o de comportamiento, elementos, que a
diferencia de los utilizados por otras técnicas, siempre acompañan al individuo.
En este área, la utilización de la voz humana como rasgo presenta un
conjunto de características que la hacen especialmente practica y la convierten en
la mejor opción, cuando no la única, en un amplio conjunto de aplicaciones.
El esquema general del proceso de reconocimiento define dos grandes
etapas: la extracción de la información relevante de las muestras de voz
capturadas, y la comparación de dicha información con otra de las mismas
características previamente almacenada; comparación, esta última, para lo cual se
suele hacer uso de técnicas de clasificación provenientes del área de la inteligencia
artificial.
Dado el estado actual de los algoritmos de clasificación, parece difícil
pensar que los sistemas de reconocimiento biométrico puedan mejorar
sustancialmente sus tasas a partir de la mejora de los mismos; es necesario, por
tanto mejorar la calidad de la información que se les suministra.
En este trabajo, el autor presenta un nuevo enfoque que permite la mejora
de las tasas del reconocimiento del locutor mediante la selección de la dicha
información, proponiendo, asimismo, un sencillo algoritmo que realiza este
filtrado. Sus resultados no sólo son aplicables al diseño de nuevos sistemas, sino
que resultan útiles a la hora de mejorar las prestaciones de los que se encuentran
en funcionamiento. ---------------------------------------------
The authentication of people identity is nowadays a crucial task, since a wide variety of systems requires a reliable method either to determine or to confirm the identity of individuals. Among all the authentication methods, the “biometric recognition” has received considerable attention in the recent years mainly due to two reasons: the strong growth in demand for security applications them, both commercial and military, and the rapid development of technology supporting it. Its purpose is to determine the identity of the person based on one or more physical or behavioural traits, elements that unlike those used by other techniques, always go with the individual. In this area, the use of the human voice as a trait has a set of characteristics that make it especially practical and it becomes the best choice, if not the only available one, for a wide range of applications. The general scheme of the recognition process is defined in two mayor stages: extracting the relevant information from the captured voice samples, and matching that information to another one previously stored of the same trait; matching, the latter, for which usually makes use of classification techniques inherit from the artificial intelligence area. Considering the current state of classification algorithms, it seems hard to believe that biometric recognition systems can substantially improve their rates just by improving them, it is therefore necessary to pay attention to improve the quality of information supplied. In this document, the author presents a new approach which allows the improvement of speaker recognition rates by the selection of such information, proposing, likewise, a simple algorithm that performs this filtering. Their results are not only applicable to the design of new systems, but also are useful in improving the performance of those which are in operation.
The authentication of people identity is nowadays a crucial task, since a wide variety of systems requires a reliable method either to determine or to confirm the identity of individuals. Among all the authentication methods, the “biometric recognition” has received considerable attention in the recent years mainly due to two reasons: the strong growth in demand for security applications them, both commercial and military, and the rapid development of technology supporting it. Its purpose is to determine the identity of the person based on one or more physical or behavioural traits, elements that unlike those used by other techniques, always go with the individual. In this area, the use of the human voice as a trait has a set of characteristics that make it especially practical and it becomes the best choice, if not the only available one, for a wide range of applications. The general scheme of the recognition process is defined in two mayor stages: extracting the relevant information from the captured voice samples, and matching that information to another one previously stored of the same trait; matching, the latter, for which usually makes use of classification techniques inherit from the artificial intelligence area. Considering the current state of classification algorithms, it seems hard to believe that biometric recognition systems can substantially improve their rates just by improving them, it is therefore necessary to pay attention to improve the quality of information supplied. In this document, the author presents a new approach which allows the improvement of speaker recognition rates by the selection of such information, proposing, likewise, a simple algorithm that performs this filtering. Their results are not only applicable to the design of new systems, but also are useful in improving the performance of those which are in operation.
Description
Keywords
Reconocimiento de voz, Biometría