Research on the accuracy of Lithuanian speaker’s identification using recurrent neural networks /

One of the main tasks of this thesis is to analyze the process of speaker identification by one’s voice by classifying a set of speaker’s voice features extracted from one’s voice sample. This thesis presents methods that are used to identify a speaker by voice. The process of speaker identification consists of several stages. Based on the research in the field, it was concluded that the most popular and commonly used method for speaker feature extraction is Mel frequency cepstral coefficients and its derivatives. This method is further applied in experiments conducted using a Lithuanian speaker dataset in order to determine a classifier most accurately identifying Lithuanian speaking individuals. The performed experimental research allowed concluding that for a Lithuanian speaker higher identification accuracy is achieved by using a recurrent neural network classifier with long short-term memory topology.

Grantor dc:publisher

Institutional Repository of Vilnius University

Year dc:date

2018

Author dc:creator

• Dovydaitis, Laurynas,

Contributors dc:contributor

• Rudžionis, Vytautas Evaldas

• speaker identification ; neural networks ; Lithuanian speaker identification

Statement dc:rights

• info:eu-repo/semantics/openAccess

Language dc:language

eng

Repository record dc:identifier

https://repository.vu.lt/VU:ELABAETD31276456&prefLang=en_US

OAI identifier oai:identifier

oai:vu.lt:elaba:31276456