Train schedule? Consult machine

2003/10/01 Andonegi Beristain, Garazi - Elhuyar Zientziaren Komunikazioa Iturria: Elhuyar aldizkaria

• Komunikazioak
• Unibertsitateak

Robots, talking machines, answering machines that understand what we have said... will be very common in our life shortly. In fact, the University of the Basque Country is developing a system capable of knowing speech. In this research they work with the universities of Zaragoza and Valencia and want to develop a machine that automatically responds to the requirements of railway information.

The machine must be able to recognize the voice of the person requesting the information, understand what was said, search for data and give an answer, for which she will also create voice.

Knowledge of speech

The first step is knowing speech. That is the work of the UPV team. To do this, first the collection of all the phonemes of the language must be completed, in this case the collection of phonemes of Castilian.

To know the train schedule, a phone call will be made shortly and a machine will assist us.

The following is information on pronunciation and co-articulation. In fact, a phoneme is pronounced differently according to the one that comes before or behind, and this is called coarticular. In addition, possible variants are taken into account depending on the context. Finally, with all this a phonemes model is built.

The second step complements the linguistic model, which is used in this context. And if someone has to ask for information about trains, he will surely use some words and some types of phrases. That is, the context reduces the richness of the language, so it is not necessary to take into account, for example, the aeronautical words, since they will never be used in the questions.

Then the phonemes model and the language model are unified and, in addition, the characteristics of improvised speech are added, such as information to understand the gaps we make from word to word, meaningless sounds, repetitions, etc.

With all these models they have developed a computer program at the UPV. This program, taking into account a statistical analysis of the voices collected in the samples and the ways of asking, interprets what the user demands in each case and therefore understands speech. This is the first step, understanding the question of using machines.

Subsequently, the UPV/EHU system integrates with the other units, that is, with which it manages the conversation, searches for the requested data, generates response, synthesizes the voice and sends information. In fact, the units that perform these functions are not shared but distributed between Zaragoza, Valencia and Leioa. As a result, large amounts of information are sent from one to another and all in real time.

Therefore, the system must be safe for fast transport of information. In addition, the separation of units has another positive feature: it is a modular system. In this way, if at any time any unit suffers any error, it is sufficient to repair or modify that unit.

• Project title Dialogue
  system for information on speaking trains.
• Develop a machine-person
  dialogue system for the use of Public Objectives.
• Principal Investigator Mª
  Ines Torres Barañano.
• UPV Template:
  A. Varona, L.J. Rodríguez, V. Gijarrubia, A. Casillas, J. Mª. University of
  Zaragoza. Polytechnic University of
  Valencia.
• Department
  of Electricity and Electronics.
• •
  
• •