Intelligence plus telephone – auto attendant and others

This is intelligent telephone technology based on speech recognition and synthesis. Simply speaking, this is an automatic telephone, which you can talk to and which can find out answers to your questions and is able not only to listen but also to hear. It is unable to be rude on the phone, to fling the receiver or to neglect its duties in general.

“The core of our technology is the speech recognition system, says Nina Petukhova, senior staff scientist, one of the development authors. In principle, this is a commercially available product already, which we did not create but only used. Our part is the software and appropriate hardware, which would enable to carry on a dialogue between a person and a machine, the dialogue being in the defined area. Certainly, we should provide the machine with a database containing responses to the asked questions or we should teach the machine to obtain the data in the interactive mode. This was not easy, but we have done that for a number of directions, and we still continue working on some projects.”

How does such a conversation take place? On the whole, it is like a conversation with a human operator, but with a very patient one. For instance, we need to book a ticket. The automatic machine will ask the caller about the date and time the ticket is needed for, it will ascertain the preferred options, it will offer the available options, if the machine fails to understand – it will repeat the question or phrase the question in other words, it will offer to choose one of available variants and so on. At that, this is not an answering machine, which is operated according to the telephone tone mode searching principle when all responses are recorded in advance. In this case, the automatic machine does recognize speech by revealing key words that contain the information necessary for an adequate answer. Once the machine received the required response (for example, the customer informed about the departure date) – it would pass to the next question – for instance, it would ask about the flight and then find out if the customer wants an aisle or a window seat.

Of course, the question arises – what about the unique individuality of a customer, or more precisely – of his/her voice? One person would speak in a deep voice, another – would squeal, but the third customer would groan like elderly people do. Will such diversity impact the speech recognition capability? It has turned out that it won’t. The system does not depend on individual speech peculiarities and it disregards them except for two special cases.

If the speech is totally indistinct, with serious logopaedic problems, the automatic operator may fail in this case and it would offer the customer to contact a human operator. But in the other case, the individuality will on the contrary be strictly accounted for and validated. For example, if the account status is to be cleared up by phone, the individual approach will be needed – first the machine will be taught by giving a reference pattern of your voice to it, and then, after the identification procedure is fulfilled, the machine will talk and answer the questions.

Of course, the authors understand that alternative variants to such “information self-service” are available now – via the Internet and simple “human” enquiry service – which operates and is sometimes very successful. However, the Internet access is not always/everywhere available, and human telephone operators get tired, fall ill and, besides, few enquiry services operate by night and on days off. In this case, speech servers will help, which are being developed by the Institute of Management Problems, Russian Academy of Sciences. Or more precisely, automated queuing systems, which do not force out the existing systems but extend them and consequently – our opportunities as well.