Information system for converting audio in Ukrainian language into its textual representation using nlp methods and machine learning

Speech recognition involves various models, methods and algorithms for analysing and processing the user’s recorded voice. This allows people to control different systems that support one type of speech recognition. A speech-to-text conversion system is a type of speech recognition that uses spoken data for further processing. It also provides several stages for processing an audio file, which uses electroacoustic means, filtering algorithms in the audio file to isolate relevant sounds, electronic data arrays for the selected language, as well as mathematical models that make up the most likely words from phonemes. Thanks to the conversion of speech to text, people whose professions are closely related to typing a large amount of text on the keyboard, significantly speed up and facilitate the work process, as well as reduce the amount of stress. In addition, such systems help businesses, because the concept of remote work is becoming more and more popular, and therefore companies need tools to record and systematize meetings in the form of written text. The object of the research is the process of converting the Ukrainian-language text into a written one based on NLP and machine learning methods. The subject of the research is file processing algorithms for extracting relevant sounds and recognizing phonemes, as well as mathematical models for recognizing an array of phonemes as specific words. The purpose of the work is to design and develop an information system for converting audio Ukrainian-language text into written text based on the Ukrainian Speech-to-text Web application, which is a technology for accurate and easy analysis of Ukrainian-language audio files and their subsequent transcription into text. The application supports downloading files from the file system and recording using the microphone, as well as saving the analysed data. The article also describes the stages of design and the general typical architecture of the corresponding system for converting audio Ukrainian-language text into written text. According to the results of the experimental testing of the developed system, it was found that the number of words does not affect the accuracy of the conversion algorithm, and the decrease in percentage is not large and occurred due to the complexity of the words and the low quality of the microphone, and therefore the recorded file.