|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ИНХС РАН |
||
Practical evaluation and validation of methods for automatic speaker identification (as applied to various languages)статья
- Авторы: Potapova R., Potapov V., Kuryanova I.
- Журнал: Lecture Notes in Computer Science
- Год издания: 2024
- Аннотация: The importance of using the results of phonoscopic examination in criminal proceedings is increasingly increasing: a voice, like a fingerprint, is individual; therefore the recognition of two voices as identical is indisputable evidence. Identification of a set of speaker’s identification characteristics, differentiation of group and individual characteristics with due account for sociolinguistic parameters, assessment of the stability of these characteristics and the degree of their expression requires the use of a complex of linguistic and acoustic methods for studying voice and speech. With this in mind, audio recordings of negotiations between ethnic criminal groups in languages, the phonological systems of which differ significantly from Russian, involve a certain difficulty, since the speech feature matrix traditionally used presents limitations for conducting identification studies in a language other than Russian. The problem of forensic research of foreign speech in a language unfamiliar to the expert was considered by us in previous papers devoted to the forensic identification of foreign-language speakers involving the methods of structural and melodic analysis of phonograms and analysis of formant trajectories of the speech signal. This paper is concerned with the consideration of fully automatic methods for identifying speakers as applied to various languages.
- Добавил в систему: Потапов Всеволод Викторович