January 17
A simulator based on neural networks will help improve pronunciation in English

Students at St. Petersburg State University of Aerospace Instrumentation have developed a simulator for correcting pronunciation in English using modern neural networks
The simulator allows users to correct pronunciation using modern neural network models, such as Wav2Vec2, pre-trained for the task of phoneme recognition. This approach ensures high accuracy of speech recognition and clear transcription. The user can record their pronunciation, receive its transcription and compare it with the reference version. The program analyzes errors, makes recommendations, and tracks learning progress. The system converts the user’s spoken speech into a digital format by analyzing acoustic signals. For this purpose, automatic speech recognition technologies are used, such as models trained on large amounts of data to recognize phonemes, words, and phrases.