Speaker identification method, involves determining statistical distribution of extracted portions of speech signal, and determining threshold value for classification of speaker by using determined statistical distribution

摘要

The method involves receiving a speech signal of a speaker, and extracting speaker specific features from the speech signal. A similarity measure is determined for correlation of the extracted specific features and reference features. Portions of the speech signal are extracted, where the speaker does not speaks in the portions. A statistical distribution of the extracted portions of the speech signal is determined. A threshold value for classification of the speaker is determined by using the determined statistical distribution. An independent claim is also included for a device for identification of a speaker.