Investigation of calculation methods of effective duration on autocorrelation-based Chinese speech recognition

Abstract

The factors extracted from running autocorrelation function (ACF) have phonetic and syllabic meanings, and they can be used for Chinese speech recognition. One of the factors is the effective duration of ACF envelope τe, which reveals the temporal repetitive continuity of the speech signals. The values of τe vary when different calculation methods are adopted. By adopting different calculation methods to several sets of Chinese vowels and characters, which were recorded from a group of speakers, different values of τe were obtained. Computed results were analyzed by combining with other ACF-based factors. © 2013 Acoustical Society of America.