Artificial intelligence

Abstract: In the article, principal technologies used in the creation of speaker identification systems and difficulties faced by their developers are considered. The structure of text-independent speaker identification using automatic segmentation of speech signal with simultaneous speaker-independent classification of segments is proposed. This approach improves accuracy of the speaker model and eliminates disagreement between training and recognizable context.

Keywords: speaker identification, wide phonetic classes, speaker model, Gaussian mixture model, speech signal segmentation, clustering of phonemes

References: