Quantifying and evaluating the impact of prosodic differences of foreign-accented English

The identification and correction of prosodic deviations in second-language speech still poses a significant challenge for computer-aided language learning. With this ultimate goal in mind, the current study compares utterances by Cantonese speakers of Canadian English with those of native English subjects through both acoustic analysis and perceptual evaluation. We aim to find measurable prosodic differences accounting for the perceptual results. Our outcomes indicate, inter alia, that unstressed syllables are relatively longer compared to stressed ones in the Cantonese corpus than in the Canadian English corpus. Furthermore, the correlations of syllabic durations in utterances of one and the same sentence are much higher for Canadian English subjects than for Cantonese speakers. The latter use a similar range of F0, but produce more and longer pitch-accents than Canadian English speakers. In a perception study we found that applying native durations together with F0 contours to the foreign-accented speech led to significantly improved listener judgments of prosodic goodness. Adjustments to duration alone also tended to yield better ratings, though the effect was not statistically significant. When durations of native English utterances were adjusted to those of Cantonese speakers, significant decrements in ratings were observed.