Speech annotation system for the comparative analysis of pronunciation in different dialects

Abstract

The article describes the development of an annotation system for the comparative analysis of pronunciation in different dialects of the Russian language. It emphasizes that voice input is nowadays widely used. The main challenge still facing voice assistants is the need to recognise different dialects. The most popular systems for the recognition of accents and dialects of Romance and Germanic languages were analysed. The key differences between speakers were determined. The article shows that phonetic features are informative enough to be used for distinguishing dialects. It also demonstrates that a large number of dialects in the Russian language requires the creation of an acoustic model for each of them. The article focuses on the analysis and modelling of the prosodic structure of the Russian dialects. The prosodic approach is based on the stress theory, which considers syllables from the point of view of emphasis and duration. It is noted that the dialects of the Russian language differ significantly in terms of their prosodic structure, i.e. their rhythmic structure, the tempo of speech, and the length of vowels. Software was developed for speech annotation which allows the user to interact with the speech database. The interface of the developed system is demonstrated. Audio recordings are visualized by oscillograms. The primary and secondary entities used in the database are highlighted. Each entity stores certain information regarding the phonemes and the speaker. The developed system for the recognition of dialects of the Russian language can be used to create a speech corpus, which will allow obtaining information about the pronunciation of phonemes using specific parameters.