‘Neural Networks Can Provide Assessments As Accurate As Humans’
Voice assistants have become part of everyday life. They can plan routes, play music and films, and answer questions. But the quality of their speech requires assessment. To address this, students of the Applied Artificial Intelligence Workshop at the HSE University and VK Engineering and Mathematics Schoolhave developed neural networks capable of evaluating speech synthesis.
Nikita Shevtsov
‘Text-to-speech technologies convert text into speech. They are the voices we hear in satnavs andaudiobooks. Previously, in order to assess how well synthesis worked, manual checking was required: people listened to recordings and gave ratings. We decided to automate this process. The task for students was to create a system that would replace subjective evaluation with a neural network model. Manual labelling is not only costly and time-consuming, but also requires the involvement of numerousspecialists,’ said Nikita Shevtsov, head of the Applied Artificial Intelligence Workshop project at the HSE and VK Engineering and Mathematics School.
The students of the workshop used open English-language SOMOS datasets containing 20,100 audio clips, 200 TTS systems, 987 assessors, and more than 350,000 ratings.
On this basis, two metrics were prepared and implemented: MOS (Mean Opinion Score)—a rating of a single audio file from 1 to 5; SBS (Side-by-Side)—a comparison of two audio clips and subsequentselection of the better one. Five models were also developed: for MOS—MOSNet, MOSNetBert, and WhisperBert; for SBS—NeuralSBS and NeuralSBSBert. The models work both in audio and multimodal formats.
‘We found that neural networks can provide assessments just as accurate as those of humans. The MOS models showed an RMSE value of ≈ 0.4, close to the human error rate of 0.62. The NeuralSBS model identifies the better audio in 73% of cases—comparable to the average subjective listener. This paves the way for faster and cheaper evaluation of TTS quality,’ noted Nikita Shevtsov.
The next step is retraining the models in Russian. The team is also considering integrating the evaluation models into CI/CD pipelines in order to implement automated quality checks in speech generation production processes.
‘Automatic evaluation of speech synthesis is a step towards more reliable and scalable TTS systems. We plan to make it accessible for wider development,’ added Nikita Shevtsov.
See also:
‘Working with AI Solves a Wide Range of Engineering Problems’
Artificial intelligence is a working tool based on a balanced combination of algorithms and engineering. Experts and doctoral students from the HSE Moscow Institute of Electronics and Mathematics explain how AI technologies can improve an application, device, or system, and what engineering tasks are solved in the process.
A New Section on AI and a Prizewinning Paper: Early-Career HSE Researchers Take Part in IEEE EDM Conference
The 27th IEEE International Conference of Young Professionals in Electron Devices and Materials (EDM) has taken place in the Altai Republic. This year, researchers from HSE University presented the results of their research and were involved in organising a new section on artificial intelligence. A paper by HSE master’s student Rodion Sidorenko was awarded third place in the research paper competition at the conference.
‘AI Enables Researchers to Tackle More Complex and Important Problems’
In late July 2026, Dmitry Rybin, a graduate of the HSE Faculty of Mathematics who is now working in China, used ChatGPT to disprove a longstanding mathematical hypothesis. In an interview with the HSE News Service, he discussed AI's ability to make discoveries in mathematics, reflected on his time at HSE University, and spoke about his doctoral research at the Chinese University of Hong Kong.
AI for Doctors: HSE Faculty of Computer Science Delivers Course for Russian University of Medicine Students
In June 2026, the HSE Faculty of Computer Science (FCS) completed a course on the use of artificial intelligence in medicine for first-year Paediatrics students at the Russian University of Medicine. The course was delivered with support from a grant awarded to HSE University under the Artificial Intelligence federal project, part of the national project ‘Data Economy and the Digital Transformation of the State.’
HSE University to Launch New AI Supercomputer
HSE University is preparing to launch its second supercomputer. The new cluster will be primarily dedicated to artificial intelligence (AI) workloads and will complement the existing cHARISMa supercomputer. It is scheduled to become operational by the end of 2026.
HSE & VK Engineering and Mathematics School Showcases 13 Projects at 10th Demo Day
The 10th Demo Day of the Joint HSE & VK Engineering and Mathematics School was held at the VK Moscow office. Students of the three workshops presented the results of 13 projects in the fields of artificial intelligence, information security, and digital platforms. Students worked on the development of recommendation services, systems for psycholinguistic text analysis and speech processing, methods for identifying celebrities in videos, algorithms for determining the toxicity of memes, security mechanisms for AI systems, and approaches to improving the effectiveness of neural network models.
Tabular Data Anonymisation Solution for Safe Use in AI Systems Developed at HSE University
The AI and Digital Science Institute at the HSE Faculty of Computer Science has developed a tabular data anonymisation service designed to prepare corporate datasets for use in analytics and AI applications. The solution can identify personal data in structured datasets, apply consistent and reproducible anonymisation rules, and generate the artifacts required for quality control, auditing, and subsequent use of data in secure environments.
HSE Scientists Develop Method to Compress Large Language Models Without Losing Quality
Researchers from the AI and Digital Science Institute at the HSE Faculty of Computer Science have developed a new compression method for large language models such as GPT and LLaMA that reduces their size by 25–36% without additional training or significant loss of accuracy. This is the first approach to use mathematical transformations—specifically, rotations of model weights—to make models more amenable to compression with structured matrices. The study results have been published in ACL Findings 2025. The code is available on GitHub.
HSE Scientists Train Neural Network to 'Hear' Faults in Electric Motors
Researchers at the AI and Digital Science Institute of the HSE Faculty of Computer Science have developed a new method—the Signature-Guided Data Augmentation (SGDA) framework—that achieves 99% accuracy in motor fault detection and 86% accuracy in fault classification. The application of this approach can reduce industrial equipment repair costs, minimise downtime, and improve production safety. The study results have been published in Engineering Applications of Artificial Intelligence.
HSE Graduate’s AI Project Wins at TECH & AI Awards
Daria Davydova, graduate of the HSE Graduate School of Business and Head of the AI Implementation Unit at the Artificial Intelligence Department of Alfa-Bank, received a prize at the TECH & AI Awards. She was awarded for the best AI solution for optimising business processes. The winners were determined as part of the VII Russian Summit and Awards on Digital Transformation (CDO/CDTO Summit & Awards).