Test environment for the two level model of Germanic prominence

In this work we present a test bed designed to verify the two level model of Germanic prominence. We give an introduction to the linguistic background of the model and derive the features that the test environment should possess. Finally we describe the details of the implementation in the ESPS/xwaves 1 environment. The implementation is based on resynthesis using PSOLA algorithm. As a linguistic application the Tone Sequence Model (TSM) is implemented and tested. 1. LINGUISTIC-PHONETIC BACKGROUND Recent developments in the theory of prominence [1][3] rigorously restrict the number of categories upon which prominence relations may be expressed. Of the number of prosodic constituents proposed in the early 80´s (see [4] for an overview) practically only two could be validated in experimental studies: the stress foot (the basic linguistic category at the level of word prosody [word stress]), and the into-national phrase (the basic linguistic category at the level of phrasal prosody). In Germanic languages the evidence that has been recently accumulated suggests that the primary acoustic correlate for prominence at the stress foot level is duration of the designated syllable [5], whereas at the intonational phrase level it is the fundamental frequency categorized as pitch-accent. It has also been observed that these two levels of prominence are fairly independent [6]. If a syllable is accented, it will necessarily be much longer than the syllable in a non-accentable position. However, if a syllable is associated to the pitch-accent within the phrase, other aspects of its production (like duration) may be, but do not have to be affected. This state of linguistic phonetic research calls for a new computational model, in which the prosody implementation is parallel and modular rather than hierarchical and incremental. 2 The two level model for Germanic prominence predicts pitch movements for stress on the phrase level, and modifications of the phone durations on the word level. A resynthesis tool that could be used to investigate the theory should give us free hand in modifying the pitch and the duration of pre-recorded speech data. The pitch synchronous overlap & add (PSOLA) algorithm as described in [10] meets our claim of independent variation of pitch and duration. Moreover , the intensity of the speech signal can be modified by a simple amplitude scaling of the output. We describe the algorithm only in few words here and suggest [10] for further reference. First the speech signal to be resynthesized …