Usage of Speech Signal Segmentation for the Construction of Complex Model in the Speaker Identification System.

The article is devoted to development of a complex speaker model for using at the text-independent speaker identification. The complex speaker model is based on gaussian mixture method. The model is formed by preliminary segmented speech signal, where each segment matches to certain broad phonetic class. Method of speaker models structuring is proposed. Speaker models are structured as a tree, which allows to identify speaker without running a full search on the set of models. Researches have shown the division of the acoustic space of speaker's voice on the set of classes that represent some phonetic events, increases the efficiency of voice identification and the proposed structuring method of models accelerates the search operation.