Cross-entropic comparison of the effects of accent, speaker and database recording on spectral features of English accents

摘要

This paper investigates the use of cross-entropy information measure for quantification and comparison of the impact of the variations of accents, speaker groups and recordings on the probability models of spectral features of phonetic units of speech. Cross-entropy measure can be used in applications such as accent identification, improved speech recognition, cross-accent phonetic-tree analysis and analysis of the influence of accents on different sets of speech parameters and models. For the purpose of this study the focus is on British English, Australian English and two different databases of American English accents (namely WSJ and TIMIT). Comparison of the cross entropies of formants and cepstrum features indicate that cepstrum features are less indicative of accents compared to formants. In particular it appears that the measurements of differences in formants across accents are less sensitive to different recording or databases. It is found that the cross entropies of the same phonemes across different accents (inter-accent distances) are significantly greater than the cross entropies of the same phonemes across different speaker groups of the same accent (intra-accent distances). The cross entropy measure is also used to construct cross-accent phonetic trees, which serve to show the structural similarities and differences of the phonetic systems across accents.