Applications of Dynamic Neural Networks and Hidden Markov Models to Pattern Recognition Tasks


APDYNN - EC-LA005

Keywords: neural networks, pattern recognition, speech and speaker recognition


Start Date: 1 January 94 / Status: finished / Duration: 24 months

[ participants / contact]


Objectivies and Approach

The objective is to study the feasibility of the mixed neural networks-stochastic modelling approach to solving the pattern recognition problem in cases where the input patterns exhibit a high time variability. Speech or electric load signals are typical examples of this type of behaviour. As for robot control, the training problem could be modelled by neural networks to solve the on-line identification of the system.

The work involves coordination meetings to be held at each partners location, followed by a short seminar. Central to the activities are a range of pre-doctoral exchange fellowships.

European links: ESPRIT Projects ELENA-NERVES-2 (6891) and NEURONET (8961).

Progress and Results

A number of neural networks paradigms were used for time series forecasting; concerning speaker recognition, a comparison has been made between artificial neural networks (ANN) and hidden Markov models (HMM) that showed a slight improvement in using HMM at the cost of a much higher information needed; a comparison was made between ANN and fuzzy neural networks (FNN) for spoken isolated word recognition that showed FNN to be slightly worse than ANN while the architecture is simpler; several architectures were implemented on transputers and the efficiency was extremely good.

Information Dissemination Activities and/or Exploitation

Within the project, 20 international publications have been made (a list of publications can be requested from the coordinator). During the coordination meeting that took place in Santiago de Chile in June 1995, a seminar on Neural Networks, Load Prediction and Fuzzy Neurons was held at the Catholic University of Santiago.


Coordinator

Universidad de Valladolid
Departamento de Informatica
Faculdad de Ciencias
Prado de la Magdalena
E - 47011 Valladolid, E

EU Partners

Universidad de Valladolid, E
Universidade de Beira Interior, P

Non-EU Partners

Pontificia Universidad Catolica de Chile, CL
Universidade de Campina, BRZ
Centro de Investigacion Cientifica y Educacion Superior, MEX

CONTACT POINT

Prof. Luis Alonso Romero
Tel: +34 83 42 3162
Fax: +34 83 42 3161


INCO synopses home page INCO cooperation index INCO keyword index
INCO acronym index INCO number index INCO Projects index
All
synopses home page all keywords index all acronyms index all numbers index

APDYNN - EC-LA005, May 1997


please address enquiries to the ESPRIT Information Desk

html version of synopsis by Nick Cook