An articulatory-functional approach to modeling Persian focus prosody

This paper is an attempt to test PENTA, an articulatory-functional model, on Persian focus prosody. The test was done on a corpus consisting of utterances with different focus conditions using PENTAtrainer2, a trainable prosody synthesizer that optimizes categorical pitch targets each corresponding to multiple communicative functions. The evaluation was done by comparing the F0 contours generated by the extracted pitch targets to those of natural utterances through numerical and perceptual evaluations. The numerical results showed that the synthesized F0 was close to the natural contour in terms of RMSE (= 1.94) and Pearson’s r (= 0.84). Perceptual evaluation showed that the rate of focus identification and naturalness judgement by native Persian listeners were highly similar between synthetic and natural F0 contours.