Toward better automatic speech recognition

Abstract

Various model techniques to adapt to various speech environments without modifying the basic automatic speech recognition were developed. Statistical data mapping assumes that speech observations are generated by subsets of mutually related random sources. It is a non-linear approach and has the strength to handle non-time-invariant variations. The warped discrete cosine transform cepstrum (WDCTC) has a better performance in a 5-vowel recognition and speaker identification task.