Previous studies on temporal aspects of languages are usually based on limited data from a small number of subjects. It is difficult to know whether these findings can really represent the general temporal aspects of continuous speech, or just the speech of the specific subjects involved. Because of this difficulty it may not be appropriate to directly apply these findings to current speech technologies.
In this study, large vocabulary continuous speech databases for Mandarin and Cantonese, recorded from large populations of subjects, are used to investigate the temporal aspects of syllables in continuous speech. Our findings include the following. No obvious temporal compensation effect has been found between the syllable initial (null initial being the sole exception) and the syllable final; Cantonese syllables exhibit less variation than Mandarin, achieving better isochrony.
