Published on: 2026-07-02
Source: Novosibirsk State University –
An important disclaimer is at the bottom of this article.
TeamStartup studios of Novosibirsk State Universitydeveloped the Talk Tuner service — an intelligent assistant for overcoming the language barrier. The project became the winner of the A:START business accelerator and a prizewinnerCatalyst accelerator.The service offers users personalized oral speech training based on artificial intelligence.
Talk Tuner is a participant in the University Technology Entrepreneurship Platform of the federal project “Technologies” of the national project “Effective and Competitive Economy.” The solution is aimed at those who have basic grammar and vocabulary but need regular practice for fluent communication.
The idea is based on the professional experience of linguist Katerina Fomel, whose teaching experience exceeds 10 years. During her work, she noticed that the main problem for students is the fear of mistakes and shyness. Existing foreign neural networks are often difficult to pay for from Russia, and ordinary bots do not provide detailed error analysis. Talk Tuner creates a safe environment for practice and provides detailed feedback.
—I have been working in the field of foreign language teaching for a long time, and I was interested in integrating artificial intelligence capabilities into this process. We focused on the most common problem — the lack of conversational practice. Research conducted confirmed the relevance of our request: users value a safe environment and quality feedback in a convenient format,— said the founder of the projectKaterina Fomel.
Currently, the AI coach is implemented as a chat bot on Telegram. The working mechanism is based on a full-fledged voice dialogue: the user sends audio messages, and the neural network responds vocally on the chosen topic. The coach not only maintains the conversation but also checks the correctness of speech. The AI identifies grammatical and lexical gaps, indicating errors to the user in text form.
The technical architecture of the service combines several neural networks simultaneously. The Whisper model is used for accurate human speech recognition. The Deepgram TTS API is responsible for the AI trainer’s voice generation. Dialogue logic and error checking are handled by the DeepSeek text model. At the same time, the architecture allows developers to change neural networks depending on the learning task.
In the startup studio at NSU, the project went through the entire path from idea to the formation of a product model. The core team consists of university representatives: the founder, Katerina Fomel, is a master’s student, and the manager, Ksenia Mikhalkina, is an undergraduate at NSU.
An important milestone in development was participating in the Catalyst accelerator. The program helped the team identify the key parameters of the MVP and begin its creation.
—Initially, we only had an idea without a clear implementation concept. Now we have formed a team, defined the product vision, and started the technical implementation of the MVP. Thanks to market analysis, we clearly understand exactly which tool is needed by the modern consumer,— notedKaterina Fomel.
Currently, the chatbot has been launched, it is being used by the first testers and clients, and the first sales have been made. In addition, Talk Tuner won the A:START business accelerator, as a result of which the team received an invitation to the Academpark business incubator. In the future, the team plans to implement a deep pronunciation analysis feature, add a database of video lessons, as well as release a mobile application and a web version of the service.
Please be advised; This information is raw content obtained directly from the information source. It represents an accurate report of what the source claims and does not necessarily reflect the position of MIL-OSI or its clients.