Perception of Polish accent in a re-synthesized speech signal

Abstract

The utterance /vidzitceteras/ was spoken, in three versions, with the accent (I) on the first, (II) on the second and (IV) on the fourth syllable. Using a package of computer programs for speech processing which executes analysis and synthesis by linear prediction, the fundamental frequency parameter, or F_0, was mutually exchaned among the three versions, in two listening experiments. In further experiments, the time variations of F_0 were averaged in several ways, and replaced either by random noise or by a constant F_0 value. The listening experiments were carried out with three groups of subjects, viz. (1) German University students, (2) Polish University students, and (3) Polish research workers and technical staff. The listeners in the first group, who knew no Polish, but specialized in phonetics, placed accent almost exactly like those of the Polish subjects whose reaction were systematic. The temporal variations of the F_0 parameter had a decisive effect on the perception of accent, whilst the effect of the other parameters was extremely limited when F o was replaced by random noise. Substitution of constant F_0 made the localization of accent almost totally impossible.