Syntactic-prosodic labeling of large spontaneous speech data-bases

Abstract

In automatic speech understanding, the division of continuously running speech i n to syntactic chunks is a great problem. Syntactic boundaries are often marked by prosodic means. For the training of statistic models for prosodic boundaries large data-bases are necessary. F or the German Verbmobil project (automatic speech{to{speech translation), we developed a syntactic-prosodic labeling scheme where two main types of boundaries (major syntactic boundaries and syntactically ambiguous boundaries) and some other special boundaries are labeled for a large Verbmobil spontaneous speech corpus. We compare the results of classiers (multilayer perceptrons and language models) trained on these syntactic{prosodic boundary labels with classiers trained on perceptual{prosodic and pure syntactic labels. The main advantage of the rough syntactic{prosodic labels presented in this paper is that large amounts of data could be labeled within a short time. Therefore, the classiers trained with these labels turned out to be superior (recognition rates of up to 96%).

VERBMOBIL: the use of prosody in the linguistic components of a speech understanding system

“…Nevertheless, the distinction of the nine classes was considered to be useful, because their automatic discrimination might become important in the future. Furthermore, these boundary classes might be marked prosodically in a different way; for a detailed discussion of the M labels see [8]. A more detailed account of the labeling scheme, an extension of the scheme as well as the computation of effort needed and the agreement between labellers (reliability) are presented in [7]; there, additional experiments are also described.…”

Prosodic modeling for Mandarin Chinese speech embedded with English spelling

“…A by-product of the approach is the automatic determination of prosodic-states. This is an additional advantage because prosodic labeling has become an interesting research topic (Batliner et al 1996;Wightman and Ostendorf 1994) and such kind of prosodic information provides cues for resolving syntactic ambiguity in automatic speech understanding (Hsieh et al 1996;Wang et al 1999) and for improving the naturalness of TTS (Chou et al 1997;de Tournemire 1997).…”

Chinese Utterance Segmentation in Spoken Language Translation

“…Linguistic analysis is performed on a shallow level. Batliner (1996) proposed a syntactic-prosodic labeling scheme in which two main types of boundaries and certain other special boundaries are labeled for a large VERBMOBIL spontaneous speech corpus. The method only aims at segmentation of these special boundaries.…”

“…Some of these approaches detect boundaries by analyzing the acoustic features of the input utterance, such as its energy contour, the speaking rate, and the fundamental frequency F 0 (Swerts 1997, Wightman 1994. It is true that some of the approaches take into account the linguistic content of the input utterance (Batliner 1996, Stolcke 1996 to some degree. For instance, automatic detection of semantic boundaries based on lexical knowledge as well as acoustic processing has been proposed (Cettolo 1998).…”