Neural Network in Russia Learns to Read Scientific Books Aloud
Researchers have developed a system that converts scientific texts into audio within minutes.
Specialists from the Artificial Intelligence Research Center at Novosibirsk State University have created a technology capable of transforming scientific literature into audiobook format. The service developed at NSU converts academic texts into audio files. The system works with PDF documents, extracts the text, processes it and generates a finished audio track.
The entire process runs automatically. The neural network analyzes the structure of a document, removes technical elements from the text and prepares the material for speech synthesis.
According to the university, the technology operates much faster than traditional audiobook recording. Converting text into audio takes roughly sixteen times less time than recording with a professional narrator.
Thousands of Audiobooks
The project is currently in the testing phase. During the trial period, the system has already produced audio versions of the first hundred scientific publications.
Developers say the technology expands ways to access scientific information. The audio format allows people to engage with academic literature while traveling or, for example, on a treadmill. It is also valuable for people with visual impairments or for listeners who prefer audio content.
The next stage of the project involves scaling the system. The team plans to convert the entire electronic library of Novosibirsk State University into audio format. The collection contains about 7,000 publications, including books published by the university press.
After the pilot phase is completed and user feedback is analyzed, the service will be offered to other libraries and educational platforms.
The “Kappa” Framework
The technological foundation of the service is the Kappa framework. It was also developed by specialists from the NSU Artificial Intelligence Research Center. The framework manages complex models and monitors the reliability of neural network operations.
According to the developers, the entire university library collection could technically be converted into audio in about a month. However, preparing the materials and verifying the results will take considerably longer. This stage alone may take up to a year.
University officials emphasized that all work is carried out in compliance with copyright law. Developers obtain permission from rights holders before processing books.
