Automatic Detection of Dialectal Features of Pskov Dialects in the Speech of Native Speakers

In this work, we made an attempt to solve the problem of detecting dialectal features in the speech of informants who have dialectal features characteristic of the Pskov dialects found on the territory of the Opochetsky district of the Pskov region and the Zapadnodvinsky district of the Tver region. The task is divided into two parts: speech recognition and features detection. First of all, we developed a model, the functionality of which include transcription of the interview collected during expeditions to the Pskov dialects. In order to find the most suitable architecture for the task, we compare several recently proposed systems. The obtained transcriptions of the interview can be used to update and expand the dialect corpus. The next step was to develop a system that can help the researcher pay attention to possible dialectal features in the annotation. The study considered several approaches to detecting dialect features, and identified the advantages and disadvantages of each. Ultimately, we developed a unified algorithm that incorporates the best of the approaches considered, it takes a .wav file as input and returns .TextGrid and .eaf files with annotations and detected dialect features.