The speech signal presents high intra and inter-individual variability, and it is also commonly contaminated with noise from various sources, non-stationary and generated by nonlinear systems. Because of these issues, automatic recognition systems or diagnosis require a processing step that will bring or remove the useful information that is implicit in a signal. Conventional signal processing and feature extraction techniques have serious limitations in capturing relevant and discriminative information from this type of signals. Thus, a demand has emerged for new intelligent algorithms to obtain information from such signals. The strategy proposed in this project is the development of novel methods of processing and analysis based on computational intelligence tools. Particularly, meta-heuristic search algorithms, deep learning networks and auto-encoders can be useful in developing improved speech features. The results of this project will be useful to exploit the information related to the emotional state of the speaker in human-machine interfaces, in order to improve the performance, the user experience and efficiency of the communication.
PICT 2014 ANPCyT
CAID 2016 UNL