Members

BEST Paper Award RTTH 2017 - Omid Ghahabi y Javier Hernando

admin — Tue, 28 Nov 2017 07:58:38 +0000

Congratulations Omid Ghahabi y Javier Hernando - Best Paper Award 2017 de la RTTH

Deep Learning Backend for Single and Multisession i-Vector Speaker Recognition

Best Paper Award RTTH 2017

Author: Omid Ghahabi and Francisco Javier Hernando Pericás

Date: 2017-04-01

Abstract: The lack of labeled background data makes a big performance gap between cosine and Probabilistic Linear Discriminant Analysis (PLDA) scoring baseline techniques for i-vectors in speaker recognition. Although there are some unsupervised clustering techniques to estimate the labels, they cannot accurately predict the true labels and they also assume that there are several samples from the same speaker in the background data that could not be true in reality. In this paper, the authors make use of Deep Learning (DL) to fill this performance gap given unlabeled background data. To this goal, the authors have proposed an impostor selection algorithm and a universal model adaptation process in a hybrid system based on deep belief networks and deep neural networks to discriminatively model each target speaker. In order to have more insight into the behavior of DL techniques in both single- and multisession speaker enrollment tasks, some experiments have been carried out in this paper in both scenarios. Experiments on National Institute of Standards and Technology 2014 i-vector challenge show that 46% of this performance gap, in terms of minimum of the decision cost function, is filled by the proposed DL-based system. Furthermore, the score combination of the proposed DL-based system and PLDA with estimated labels covers 79% of this gap.

Reference: Ghahabi, O., Hernando, J. Deep learning backend for single and multisession i-vector speaker recognition. "IEEE-ACM Transactions on Audio Speech and Language Processing", 1 Abril 2017, vol. 25, núm. 4, p. 807-817.

Link: http://ieeexplore.ieee.org/document/7847321/?reload=true

UPCommons: http://upcommons.upc.edu/handle/2117/104282

The other finalists were:

--------------------------------------------------------------------------
Emilio Granell and Carlos-D. Martínez-Hinarejos "Multimodal Crowdsourcing
for Transcribing Handwritten Documents"
--------------------------------------------------------------------------

Gustavo Andrade-Miranda, Nathalie Henrich Bernardoni, Juan I.
Godino-Llorente "Synthesizing the motion of the vocal folds using optical
flow based techniques"
------------------------------------------------------------------

Deep Learning, Speaker Recognition

TALP Welcome to Magdalena Biesialska PhD Student

admin — Thu, 23 Nov 2017 12:15:58 +0000

TALP Welcome to Magdalena Biesialska PhD Student

Magdalena Biesialska has started his PhD at TALP under the advice of Dra. Marta Ruíz Costa-jussà and Dr. Adrian Rodríguez Fonollosa.

She will research on learning multilingual vector representations for neural machine translation.

MT, Deep Learning, PhD student

CIOReview: Herta Security Winners of 20 most promising video surveillance companies of 2017

admin — Wed, 11 Jan 2017 07:31:18 +0000

Congratulations Javier Hernando on Herta Technologies TALP spinoff latest awards.

In January 2017 CioReview recognized Herta as one of the 20 most promising video surveillance companies of 2017, after being awarded in December 2016 as the world's top 100 technology companies by Red Herring Global 2016.

spin-off, herta

Dr Javier Hernando new TALP director

admin — Sun, 07 Aug 2016 22:00:00 +0000

Javier Hernando received the M.S. and Ph.D. degrees in telecommunication engineering from the Technical University of Catalonia (UPC), Barcelona, Spain, in 1988 and 1993, respectively. Since 1988, he has been in the Department of Signal Theory and Communications, UPC, where he is currently a Full Professor and the Director of the Research Center for Language and Speech. During the academic year 2002–2003, he was a Visiting Researcher in the Panasonic Speech Technology Laboratory, Santa Barbara, CA, USA. He has led the UPC team in several European, Spanish, and Catalan projects and he has also been involved in technology transfer projects. His research interests include robust speech analysis, speech recognition, speaker verification and localization, oral dialogue, and multimodal interfaces. He is the author or coauthor of about two hundred publications in book chapters, review articles, and conference papers on these topics. He received the 1993 Extraordinary Ph.D. Award of UPC.

In 2009, Javier Hernando founded Herta Security, S.L. to generalize voice biometrics to other biometric modalities, such as voice, face and iris, and nowadays it is focused on biometric video-surveillance. During last years, TALP students have joined Herta, and both TALP and Herta have collaborated in joint research projects. In 2014, BioFinder, Herta’s product to track faces in long and highdefinition recording, was the recipient of the Security & Fire Excellence Award 2014 as the best system of the year (Excluding Cameres and Lens). This award sponsored by IFSEC International and FIREX International.

GE awards winners of Flight Quest 2!

Super User-2 — Thu, 24 Apr 2014 11:38:38 +0000

GE has revealed the private leaderboard on Flight Quest 2 -- and four outstanding winners! The competition has been running since August as the second part of GE's Industrial Internet Flight Quest with a challenge to develop algorithms that increase flight efficiencies in real time, reducing delays and maximizing a flight's profitability. Using national airspace data provided by FlightStats, the winning algorithms determine the most efficient flight routes, speeds, and altitudes by taking into account variables such as weather, wind, and airspace restraints. The 1st place model by José Fonollosa proved to be up to 12 percent more efficient when compared to data from actual flights.

Flight Quest 2 was another massive step in GE's innovative approach to imagining the future of aviation. With some 3800 submissions to the leaderboard, Kagglers simulated well over 50,000,000 flights by modelling their waypoints into the scoring Simulator. Our hats are off to the hard work of everyone on the leaderboard… this is one of the most challenging problems we've ever hosted.

Jose A. R. Fonollosa has a M.S. and Ph.D. in electrical engineering from the Universitat Politëcnica de Catalunya (UPC, 1986 and 1989). He is a member of the Department of Signal Theory and Communications since 1986, currently as Full Professor, and visitor, from August 1991 to July 1992, of the Signal and Image Processing Institute, University of Southern California. Jose received the 1992 Marconi Young Scientist Award. He was co-founder (1999) of the company VERBIO S.L., specialised in spoken language technology. From 2006 to 2010, he was director of the Center for Language and Speech Technologies and Applications (TALP). He has published more than 150 scientific papers in statistical signal analysis, artificial intelligence, machine learning and optimization.

IberSPEECH 2012 Microsoft Best Demo Award

admin — Fri, 03 Feb 2012 23:00:00 +0000

Search Engine for Multilingual Audiovisual Contents

Search Engine for Multilingual Audiovisual Contents

Authors: José David Pérez, Antonio Bonafonte, Antonio Cardenal, Marta Ruiz, José A. R. Fonollosa, Asunción Moreno, Eva Navas, Eduardo R. Banga

Abstract: This paper describes the BUCEADOR search engine, a web server that allows retrieving. multimedia documents (text, audio, video) in diferent languages. All the documents are translated into the user language and are presented either as text (for instance, subtitles in video documents) or dubbed audio. The user query consist in a sequence of keywords and can be typed or spoken. Multiple Spoken Language Technologies (SLT) servers have been implemented, such as speech recognition, speech machine translation and text-to-speech conversion. The platform can be used in the four Spanish oficial languages (Spanish, Basque, Catalan and Galician) and in English.

Keywords: multimedia search, multilingual search, speech recognition, machine translation, speech synthesis, speech to speech translation.

Link: http://iberspeech2012.ii.uam.es/IberSPEECH2012_OnlineProceedings.pdf

Buceador, MT, ASR, TTS, multimedia search, multilingual search, speech to speech translation.

Dr Lluís Padró new TALP director

Super User-2 — Tue, 10 Jan 2012 06:11:00 +0000

Dr. Lluís Padró is Associate Professor at Universitat Politècnica de Catalunya since 1999.

His research ranges across several areas in NLP, with a main focus on basic language processing (tagging, parsing, sense disambiguation, etc). He is the creator and leader of the Freeling project, an open source suit of language analyzers relased under the GNU General Public License of the Free Software Foundation.

He has participated in several EU funded projects and Spanish government funded projects, acting as coordinator in 5 of them. Moreover, he has also been involved in several technology transfer projects.

NLP

Members