Advanced search options

Advanced Search Options 🞨

Browse by author name (“Author name starts with…”).

Find ETDs with:

in
/  
in
/  
in
/  
in

Written in Published in Earliest date Latest date

Sorted by

Results per page:

You searched for id:"handle:10486/678952". One record found.

Search Limiters

Last 2 Years | English Only

No search limiters apply to these results.

▼ Search Limiters


Universidad Autónoma de Madrid

1. López Moreno, Ignacio. Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition .

Degree: 2017, Universidad Autónoma de Madrid

Artificial neural networks are powerful learners of the information embedded in speech signals. They can provide compact, multi-level, nonlinear representations of temporal sequences and holistic optimization algorithms capable of surpassing former leading paradigms. Artificial neural networks are, therefore, a promising technology that can be used to enhance our ability to recognize speakers and languages–an ability increasingly in demand in the context of new, voice-enabled interfaces used today by millions of users. The aim of this thesis is to advance the state-of-the-art of language and speaker recognition through the formulation, implementation and empirical analysis of novel approaches for large-scale and portable speech interfaces. Its major contributions are: (1) novel, compact network architectures for language and speaker recognition, including a variety of network topologies based on fully-connected, recurrent, convolutional, and locally connected layers; (2) a bottleneck combination strategy for classical and neural network approaches for long speech sequences; (3) the architectural design of the first, public, multilingual, large vocabulary continuous speech recognition system; and (4) a novel, end-to-end optimization algorithm for text-dependent speaker recognition that is applicable to a range of verification tasks. Experimental results have demonstrated that artificial neural networks can substantially reduce the number of model parameters and surpass the performance of previous approaches to language and speaker recognition, particularly in the cases of long short-term memory recurrent networks (used to model the input speech signal), end-to-end optimization algorithms (used to predict languages or speakers), short testing utterances, and large training data collections.; Las redes neuronales artificiales son sistemas de aprendizaje capaces de extraer la información embebida en las señales de voz. Son capaces de modelar de forma eficiente secuencias temporales complejas, con información no lineal y distribuida en distintos niveles semanticos, mediante el uso de algoritmos de optimización integral con la capacidad potencial de mejorar los sistemas aprendizaje automático existentes. Las redes neuronales artificiales son, pues, una tecnología prometedora para mejorar el reconocimiento automático de locutores e idiomas; siendo el reconocimiento de de locutores e idiomas, tareas con cada vez más demanda en los nuevos sistemas de control por voz, que ya utilizan millones de personas. Esta tesis tiene como objetivo la mejora del estado del arte de las tecnologías de reconocimiento de locutor y de idioma mediante la formulación, implementación y análisis empírico de nuevos enfoques basados en redes neuronales, aplicables a dispositivos portátiles y a su uso en gran escala. Las principales contribuciones de esta tesis incluyen la propuesta original de: (1) arquitecturas eficientes que hacen uso de capas neuronales densas, localmente densas, recurrentes y convolucionales; (2)… Advisors/Committee Members: González Rodríguez, Joaquín (dir.) (advisor), González Domínguez, Javier (dir.) (advisor).

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

López Moreno, I. (2017). Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition . (Thesis). Universidad Autónoma de Madrid. Retrieved from http://hdl.handle.net/10486/678952

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

Chicago Manual of Style (16th Edition):

López Moreno, Ignacio. “Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition .” 2017. Thesis, Universidad Autónoma de Madrid. Accessed April 22, 2018. http://hdl.handle.net/10486/678952.

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

MLA Handbook (7th Edition):

López Moreno, Ignacio. “Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition .” 2017. Web. 22 Apr 2018.

Vancouver:

López Moreno I. Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition . [Internet] [Thesis]. Universidad Autónoma de Madrid; 2017. [cited 2018 Apr 22]. Available from: http://hdl.handle.net/10486/678952.

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

Council of Science Editors:

López Moreno I. Deep Neural Network Architectures for Large-scale, Robust and Small-Footprint Speaker and Language Recognition . [Thesis]. Universidad Autónoma de Madrid; 2017. Available from: http://hdl.handle.net/10486/678952

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

.