Daniel Hirst

Cross-linguistic prosodic comparison with OMProDat database

OMProDat is an open multilingual prosodic database, which aims to collect, archive and distribute recordings and annotations of directly comparable data from different languages representing different prosodic typological characteristics. OMProDat contains recordings of 40 five-sentence passages read by 5 male and 5 female speakers of each language. Currently the database contains recordings for five languages: English, French, Mandarin Chinese, Cantonese and Korean. In this study, the generation of German database is introduced, and a prosodic comparison of these six languages is described. 18 melodic metrics were automatically extracted from each language, and can discriminate these six languages with an accuracy rate of 61.88%. Moreover, the dynamic melodic characteristics of the dialect in the German database can also be illustrated with these melodic metrics.

ProZed: A Speech Prosody Editor for Linguists, Using Analysis-by-Synthesis

This chapter describes a tool designed to allow linguists to manipulate the prosody of an utterance via a symbolic representation in order to evaluate linguistic models. Prosody is manipulated via a Praat TextGrid which allows the user to modify the rhythm and melody. Rhythm is manipulated by factoring segmental duration into three components: (i) intrinsic duration determined by phonemic identity (ii) local modifications encoded on the rhythm tier and (iii) global variations of speech rate encoded on the intonation tier. Melody is similarly determined by tonal segments on the tonal tier (= pitch accents) and on the intonation tier (= boundary tones) together with global parameters of key and span determining changes of pitch register. The TextGrid is used to generate a manipulation object that can be used either for immediate interactive assessment of the prosody determined by the annotation, or to generate synthesised stimuli for more formal perceptual experiments.