Elena, "electronic girl" from the support service

Hi, Habr. Today we would like to tell you about a wonderful girl. More precisely, about the "girl". This is a virtual operator in the call center support service, and her name is Elena. She is able to recognize voice requests and respond to them.

What is it for?

Typically, callers call customer support to ask various questions. In many cases, these questions are of the same type. Once, in order not to make subscribers wait for the operator to respond, a voice menu (IVR) was invented. You all came across them: "... to contact the operator, press" zero "." However, progress does not stand still, and now there are new, more convenient technologies. One of them is the voice query recognition system, which allows you to maintain the degree of automation of support services.
')
So we introduced the virtual operator Elena. Or assistant, as you like. So far, Elena can not be attributed to a full-fledged artificial intelligence systems. At the moment it is a kind of superstructure above the IVR, redirecting the person to specific sections of the menu. But, in general, Elena is closer to such systems as Siri on the iPhone, or to mobile consultants on Android. So far, its capabilities are much more modest, and it will not be able to tell you where the nearest metro station is or how to catch on the moth. But Elena can already recognize the most popular questions and give answers to them. And in the future in general will turn into a wonderful interlocutor. But more on that below.

What is Elena able already?

So far, the system has been put into trial operation, and only customers of the Metropolitan Branch can use it. Now Elena handles typical and most requested questions. Elena's database is constantly growing, but if she cannot recognize the question or does not find the answer, then she transfers the call to the appropriate voice menu item.

To date, the system is able to recognize about 70 different issues. For example, she can answer which paid services are connected, how to set up a “favorite number”, what the subscriber’s phone number is, etc. At the same time there is no stereotyped wording. The technology of continuous speech recognition allows Elena to “understand” freely constructed phrases.

How to work with Elena?

The system operates in the "Question-Answer" mode. When Elena answers your call, she asks you just one question: “How can I help you?”. You can in a free form, like a living person, ask anything. No need to try to simplify the phrase or communicate, as with the savage: "My your talk." We teach Elena using ordinary human speech.

Helen also has an excellent ear, so speaking louder is not required. But at the same time it is better not to use the speakerphone, it can degrade the quality of recognition.

Elena can ask clarifying questions, they should also be answered the same way as an ordinary person. If Elena gave you an inaccurate answer, it is better to ask again or rephrase your question.

You can talk with Elena by calling 8-800-550-07-67.

How is Elena?

The system is based on the development of the Speech Technology Center (MDG) company from St. Petersburg, which specializes in creating speech recognition and speech synthesis systems. For example, this year, the MDG announcer identification technology won first place at the NIST's I-vector Challenge 2014 Worldwide Competition.

It is expected that the project will pay off during the year due to operator unloading. While the trial operation is underway, Elena's work is constantly monitored by a team of developers and call center employees. Currently, the percentage of correctly processed requests is twice the estimated one.

The database of the system consists of metathemes, each of which combines different queries. Adding a new query to an existing metatheme takes several minutes. But the duration and complexity of creating a new metatheme depends on a number of factors: the presence of a similar basic metatheme in the MDGs, the presence of examples of real hits on this metatheme, and so on.

Currently, system resources allow processing 1 million requests per year. After launching into commercial operation at the Stolichny Branch, Elena will process 12 million requests in Moscow. And with access to the federal level - 60 million.

Further development of the project

By the end of the year, Elena is expected to take on a substantial share of requests from subscribers of the MegaFon Metropolitan Branch. According to our estimates, as a result, this system will reduce the burden on support service operators by one third. But for this, much more needs to be done:

• Carry out further tuning of the system based on the analysis of communication with subscribers.
• Significantly expand the list of questions that the system can answer.
• Expand Elena's ability to ask clarifying questions.

Her further plans are to make the service available throughout the country, as well as to expand Elena’s functionality so that she can communicate in the support chat on MegaFon sites. Next, we want to create a full-fledged mobile application in which Elena will “live”. She will not only communicate in voice, but also be able to display various information on the screen, send e-mails.

In the long run, we want to make Helen a real person. So that it was not just a talking reference book, but acquired its own character. We will try to make Helen a real companion, with whom it will be just interesting and pleasant to communicate.

By the way, Elena already has her “face”, visual image. And the girl at the beginning of this post is a surprise! - not Elena. We are not planning to show it yet, after all there must be a mystery in a woman. But if we talk about the capabilities and nature of the virtual assistant, how would you like to see him? What would you like to ask Elena?