OFFER DEADLINE01/09/2018 13:00 - Europe/Brussels
EU RESEARCH FRAMEWORK PROGRAMMEH2020 / Marie Skłodowska-Curie Actions
ORGANISATION/COMPANYInternational Research Projects Office
DEPARTMENTPromotion and Advisory Unit
Professor Antonio M. Peinado, from the Department of Signal Theory, Networking and Communications at the University of Granada, welcomes postdoctoral candidates interested in applying for a Marie Skłodowska-Curie Individual Fellowships (MSCA-IF) in 2018 at this University. Please note that applicants must comply with the Mobility Rule (more information: http://sl.ugr.es/09Qg).
Brief description of the institution:
The University of Granada (UGR), founded in 1531, is one of the largest and most important universities in Spain. It serves more than 60000 students per year, including many foreign students, as UGR is the leader host institution in the Erasmus program. UGR, featuring 3650 professors and more than 2000 auxiliary personnel, offers a total of 75 degrees through its 112 departments and 28 centers.
UGR is also a leading institution in research, located in the top 5/10 of Spanish universities by a variety of ranking criteria, such as national R&D projects, fellowships awarded, publications, or international funding. UGR is one of the few Spanish Universities listed in the Shanghai Top 500 ranking (http://www.arwu.org/), and it is also well recognized for its web presence (http://www.4icu.org/top200/).
Internationally, we bet decidedly by our participation in the calls of H2020, both at partner and coordination. For the duration of the Seventh Framework Programme, the UGR has obtained a total of 66 projects, with total funding of 17.97 million euros, and for H2020, until 2015, more than 25 projects with total funding of more than 6 million euros. Our more than 3,000 researchers are grouped into 365 research groups covering all scientific fields and disciplines.
Brief description of the Centre/Research Group
SIGMAT, SIGNAL PROCESSING, MULTIMEDIA TRANSMISSION AND SPEECH/AUDIO TECHNOLOGIES (ceres.ugr.es)
SigMAT is a Research Group of the University of Granada, code TIC-234 (Junta de Andalucia). It was created in 2012 and it is integrated by 4 faculty members and 8 researchers. The main objectives of SIGMAT Group can be summarized in the following lines:
- Research, innovation and development of speech technologies: voice recognition, identification and speaker verification, speech coding, speech synthesis.
- Development of robust transmission techniques of multimedia information: speech, images, video.
- Research in other areas of signal processing, such as nondestructive evaluation and proteomics.
- Development of applications related to speech technology.
During the last years, several speech technologies (e.g., Google search or Vlingo) have jumped from the laboratories to the real world and have become popular with new well-known applications. This jump has been partially possible to the emergence of speech technologies which allow to tackle the problem of processing speech in adverse acoustic environments. Despite of the big effort devoted to this issue, acoustic noise still is one of the main drawbacks for deployment of some applications. For example, the state of the art in speech recognition systems achieves no more than 90-95% of accuracy in low-medium complexity tasks with a limited vocabulary. Thus, the need of a paradigm change and the importance of exploring new approaches, like those derived from machine learning theory, is clear. In this project we propose the integration of deep neural networks in multimicrophone systems for the goal of achieving a high performance in speech applications (in particular, enhancement, recognition and/or verification) operating in realistic conditions.
Regarding the connectionist approach, during the last years we have witnessed to the rise of machine learning and, in particular, DNNs in the field of speech processing. This success has been supported by the development of new training methods, as well as new computation capabilities, which has helped to overcome the problems that led to a reasonable skepticism concerning the utility of this type of approaches. However, the growing optimism created by these techniques has led to their extended use as black boxes capable to performing an integral processing, from the signal to the desired response, wasting the knowledge we could have about the problem to be solved. On the contrary, the traditional approaches based on models and algorithms are based on the a priori knowledge we have, although often require approximations to make them feasible, but that reduce their performance. This project intends the integration of both visions, taking the best of each one and exploiting the previous knowledge of the signa and its modeling, in order to develop new techniques for speech processing.
The second issue considered in this project is the multichannel context. Currently, an increasing number of devices (smartphones, tablets, multimicrophones for multiparty interaction, etc.) use more than one microphone for speech acquisition. However, when we have to deal with realistic environments (streets, airports, stations, etc.) and/or systems with few microphones (as in the case of mobile devices), classical beamforming methods cannot provide the required performance. While we can find a number of works where DNNs are applied to single-channel systems, their extension to multichannel systems (as an alternative to classical processing) is still rather unexplored. Therefore, we consider that the integration of DNNs in multichannel systems will provide new tools that will enable speech-driven applications even in very adverse acoustic environments, mimicking the high performance that can be achieved by the human auditory system.
- Information Science and Engineering (ENG)
For a correct evaluation of your candidature, please send the documents below to Professor Antonio M. Peinado (firstname.lastname@example.org):
- Letter of recommendation (optional)