New feature vector extraction method for speaker recognition

Speech signal contains information not only connected to the pronounced phrase, but also data about speaker, language, environment, emotional state of the speaker. The main objective of the research is development of methods and algorithms increasing the precision of speaker recognition preserving acceptable indicators on computational complexity. Extraction of vectors of speech signal is an important stage of speaker recognition. Method based on Hilbert-Huang transform considering instability and non-linearity of human speech, as well as effective noise cancelling of the spectrum was proposed in the article.