The project uses Xu´s PENTA model to analyze and synthesize tonal prosody (in terms of F0) in Swedish. The model is based on two principles: we speak in order to communicate and speech depends on physiological mechanisms. F0 is modeled as an approximation to targets, which are determined by communicative functions. Swedish, a pitch-accent language, will make an interesting test case for the model.
In this project we will use Xu´s PENTA model to analyze and synthesize tonal prosody in two Swedish dialects. The PENTA model is quantitative and based on two first principles: first, we speak in order to communicate, second, speech depends on physiological mechanisms. Thus, the model both describes and explains F0 patterns in utterances. F0 is modeled as the outcome of an approximation to targets, which in turn are determined by communicative functions. The process is reversible, i. e. it is possible to synthesize F0 by feeding parameter values into the model and let it calculate F0 accordingly. We will focus on three communicative functions: signaling word accent, speech act and information structure. For that, we will design and analyze spoken corpora where these functions are varied in a controlled manner. The results will be implemented in speech synthesis, where the model will be used to generate F0. Finally, both the analysis and synthesis will be evaluated by means of listening tests. The PENTA model has been applied to a tone language (Mandarin) and an intonation language (English). By testing it for Swedish, a pitch-accent language, the project will make an important contribution to the further development of the model. Moreover, we will focus on two dialects of Swedish, Standard and South Swedish, which exhibit a significant difference in the realization of focus, a case well suited for evaluating the PENTA model.