One of the leaders in the field of speech technologies built on the basis of AI is the Greek company Omilia, a developer of a voice system for intelligent recognition of natural speech. The company's solution based on the Omilia DiaManT platform uses algorithms based on the work of neural networks, which allows to achieve a human-level understanding of customer requests. The company has implemented many successful projects in more than 15 countries of the world, including the countries of our region, and now, together with its official partner in Azerbaijan, iTech Group, is preparing to implement this solution in one of the largest companies in our country. Rovshan Akbarov, Director of Business Development at iTech Group, talks about the prospects for implementing Omilia technologies in an interview with Infocity

- Rovshan, last time we talked about cybersecurity, but the development of AI is no less interesting. And one of the trends in this area, of course, are technologies that involve voice processing. How far has Omilia progressed in this direction, and what challenges does Omilia's technology meet today?

- In short, these technologies allow you to save a lot and at the same time improve the quality of service in call centers. For example, banks, telecommunications companies and government organizations process a huge number of requests daily, despite the fact that most of them are standard requests. Call center operators respond to such calls using a standard protocol, which can also be done by a voice robot. At the same time, the individual will have more time to handle specific issues and consult with priority clients. Imagine a situation when you called the bank, but did not get into the traditional IVR menu, where you need to select the desired service from a long list or wait for a connection with the operator, and you were greeted by a voice robot with a high level of speech recognition from Omilia. He can not only identify the client and understand the question, but also independently voice information upon request or redirect to the right specialist. This technology can also completely replace operators when collecting feedback from customers.

Omilia's voice-based intelligent speech recognition system provides only one open question during customer service - what information is of interest to the customer. In the future, she perceives the answers of subscribers in a free form with an accuracy of more than 95% and conducts a lively unstructured dialogue. The use of this technology will significantly reduce the cost of maintaining call centers, since the voice robot will respond to customer requests without breaks and days off, speed up data collection and delivery of information to customers, and also significantly increase customer retention rates. And there is every reason to believe that the spread of voice robots is just the beginning of a fundamental trend towards the use of AI in our country.

"But each country has its own dialects and many variations on how people use language to ask questions and interact. As far as we know, Azerbaijan does not yet have its own language model for AI...

- Each country has different levels of technology maturity, and the creation of a language model is associated with a lot of concomitant factors, among which the diversity of dialects is only one of the few. The first and main task of developers is to collect high-quality structured information for teaching a language model. In addition to the fact that not so many companies in Azerbaijan can actually collect such an extensive voice database, the creation of a model is complicated by the fact that there are no generally accepted standards by which quality could be measured. But nevertheless, the Azerbaijani language model should be ready towards the end of this year, which will allow to show all the possibilities of Omilia solutions, which are perfectly trained for specific AI tasks in almost all possible work scenarios. And I am sure that the creation of a language model in Azerbaijan will be of great help not only in the development of voice technologies, but will also be useful in many other services, for example, it will speed up the delivery of answers in search engines, improve the quality of translations, and create the ability to evaluate the content of posts in social networks. etc. In addition, the language model will also help the development of projects by local developers, who today are taking quite bold and productive steps towards creating voice robots.

- Some companies are proud that they do not use bots in their work, and only real people communicate with clients. How wrong is such an approach to building the work of call centers today?

- Just as every enterprise has a lot of repetitive tasks, so every call center has a lot of typical response templates. At the same time, when we contact a call center, we are not always happy with a long wait for an operator's response, especially if we need to quickly get a fairly simple and monosyllabic answer. In addition, some operator may simply be in no mood, tired or simply incompetent, which also does not add joy to communicating with him. Of course, many customers do not yet trust fully automated systems, assuming that they will not understand their question, but the threshold for understanding such systems is growing every day. More and more complex models are being trained, capable of behaving so naturally that at the end of the conversation, customers, especially the elderly, thank the voice robot, mistaking its answers for the speech of a real person.

And most importantly, such technologies are now available not only to large-scale enterprises, but also to companies with a small staff. And this is also directly related to the maturity of technology. Omilia began implementing its solutions with large clients in North America back in 2015, and then these were multi-level projects, the implementation of which took from 12 to 18 months and cost several hundred thousand dollars. Today, things are very different because Omilia's technology has been deployed in a lot of large enterprises, the company has trained models for each industry, so it doesn't have to do it again as soon as a new customer arrives. Now the technology is so perfect that the developer makes it available right out of the box from their own cloud platform for companies of any size. For Omilia, it doesn't matter if you have five client requests per day or thousands, you can still benefit from using this technology.

- Most companies have their own customer support infrastructure. How easily does Omilia integrate into it? Does the solution depend on the infrastructure or can customers use it anywhere?

- For Omilia, it doesn't matter what kind of infrastructure the customer uses. iTech Group, as an integrator and official partner of Omilia in Azerbaijan, can deploy the solution both locally and in the cloud for a variety of users. This company strives to provide technologies that require a minimum of implementation effort, which will not force the integrator to involve development teams for the subsequent polishing of the solution. And most importantly, the solution can be delivered today. It is so easy to use that the company's portfolio even includes examples of customers deploying additional modules for their company's needs on their own. This is encouraging as it confirms that Omilia does indeed have the technology polished to the point where customers can use it themselves. And this is a very serious topic of conversational AI, because there is a lot of hype in the voice systems market. There are too many offers and too many customers who use the services of fairly well-known manufacturers, but when they try to create something that will suit their needs, they experience nothing but disappointment. Omilia, on the other hand, gives room for imagination, a wide range of possibilities for adding new modules and scenarios.

- With the development of such systems, one can only guess what the future holds for us?

- In fact, such technologies can no longer be called "raw", but now work is underway on recognizing emotions and the emotional coloring of each separately spoken word, closer integration with video calls and determining personality, request and mood by face and lip movement. Biometric technologies are also developing, recognizing the user by voice. And in the near future, they will become so flawless in use that they can save people from routine work, perform various tasks better, faster and more efficiently.