KONUŞMACI TANIMA İÇİN ÖZELLİK SEÇİMİ VE SINIFLANDIRMA TEKNİKLERİ

Abstract

Speaker recognition can be considered as a subset of the more general area known as pattern recognition, which may be viewed basically in three stages as: feature selection and extraction, classification, and pattern matching. Extensive research in the past has been directed towards finding effective speech characteristics for speaker recognition. But, so far, no feature set is found to be known to allow perfect discrimination for all conditions. As the performance of features depends on the nature of application, the selection of salient features is a key step in the recognition process. In this paper, we present a general view of speech features and well known classifiers originally developed for text-independent speaker recognition systems. A comparative discussion on choice of suitable speech features and classification techniques is also given.

Keywords

• Özellik, Sınıflandırma, Doğrusal öngörülü kodlama, Gizli, Markov modeli, Karma Gaussian modeli