speech-interactive_main_goals

Main goals

The Speech Interactive Research Group is devoted to develop research and to transfer technology in areas related to pattern recognition and applications focusing on speech and language technologies. Our main goals are to develop:

  • Spoken dialog systems focusing on statistical dialog managers and online adaptation to the task and users.

  • Statistical language generation methods for interactive interfaces.

  • Methods to identify emotions from speech signal focusing on robust detection of emocional activation in interactive speech.

  • Techniques to classify sentiments from social media focusing on difficult emotional language such as sarcasm, nastiness etc. on dialogic context.

  • Technologies to learn from human-to-human interaction and to automatically generate interactive language.

These developments require:

  • Research on pattern recognition methods and applications as well as machine learning algorithms. Specifically we are interested in stochastic finite-state models (automata, transducers and bi-automata), grammar inference algorithms for parameter estimation, online learning algorithms for adaptation, SVM classifiers and feature optimization.

  • Development of the linguistic resources needed for statistical model inference, focusing on spontaneous interactions as well as linguistic resources for basque language. We are also aimed at using crowdsource resources and procedures.

  • Development of prototypes and demonstrator systems in collaboration with technological centers and companies as a first step to technology transfer.

To this end some additional and important group goals are

  • to advise graduation students, master thesis and PhD students as well as to train technicians that can be potentially contracted by partners and collaborators;

  • to keep close collaboration and advising procedures with technological centers and companies in our areas of expertise;

  • to collaborate with international research groups in USA and Europe.