FINCA Bank has implemented a voice artificial assistant in Kyrgyzstan, working in the Kyrgyz language. The innovative system is already functioning in the organization's contact center and is able to fully process both incoming and outgoing customer requests. Users no longer need to listen to long voice menus or memorize special commands. It is enough to formulate your question in ordinary natural speech, after which the neural network independently recognizes the request and selects the most appropriate response scenario. This solution is based on advanced speech recognition and synthesis technologies. The first technology converts the client's voice into text, and the second technology voices the response generated by the system.

According to representatives of the bank, ready-made market solutions for the Kyrgyz language did not provide the necessary quality for professional banking services. Therefore, FINCA Bank, together with its technological partners, has developed separate specialized models for the recognition and synthesis of Kyrgyz speech. During the development, all the features of the spoken language were carefully taken into account. Users can shorten words, change the order of phrases, speak at different speeds, pause, and use dialect expressions. That is why the model was trained mainly on live speech, and not only on strict written texts. After recognition, the system determines not just individual keywords, but the general meaning and true intention of the client. Different formulations of a single request can be automatically combined into a single service scenario.

If a customer needs to contact a live employee of the bank, he just needs to ask the system to switch the call to the operator. After that, the request is instantly transmitted to the relevant specialist of the contact center. The artificial assistant advises on the bank's products and services, answers the most frequent questions and, if necessary, connects a person. The bank itself notes that the system will be constantly improved as new applications accumulate. Deep analysis of dialogues should help improve the quality of Kyrgyz speech recognition and the overall accuracy of responses. This project is also of great importance for the development of voice technologies in the Kyrgyz language, since the amount of available text and audio data for its learning is significantly less than for the most widely spoken world languages.