Parameterization and automatic labeling of Hungarian intonation

Abstract

In Hungarian intonation research the goal of a common framework developed by Varga (2002; [1]) is to categorize the intonation within the domain of accent groups by character contours. We propose a linear parameterization of a subset of these contours derived from polynomial stylization. These parameters were used to train classification trees and support vector machines for contour prediction. Parameter extraction and training was carried out on the original F0 contours of spontaneous speech data as well as on three differently normalized variants suppressing fundamental frequency level and range effects. The highest accuracies were obtained for classification trees and F0 residuals after midline subtraction, but the overall performances were rather poor. Nevertheless, a significant improvement of the results was achieved by a Hidden Markov model to predict the correct label sequence from the partly erroneous classification output.

Contour clustering: A field-data-driven approach for documenting and analysing prototypical f0 contours

“…Resynthesised contours were then generated on the basis of both the global and local contours and were rated above average on their naturalness by human listeners. Results showed the potential of the fully automated procedure, including an evaluation of the semantic weight, information status and utterance finality as encoded by the individual classes (Reichel 2012).…”

Aspects of gestural alignment in task-oriented dialogues

“…The values of variables based on the measurements of these features were calculated and compared between the initial and the final stages of dialogues, between dialogue types (collaborative vs. competitive), and taking into account the nationality of the participants. The co-occurrence of gesture events and functions was understood as in the previous studies by the present authors referring to speech events (Karpiński et al, 2014) or interactions between gestures and speech (Czoska et al, 2015). Co-occurrence was inspected using the moving frame (or moving windows) approach (see also : Fig.…”

“…The moving frame method was implemented as an Annotation Pro plugin (a C# script). The initial version of the plugin (SRMA) was tested and used beforehand to study the variability of speaking rates in two different corpora of task-oriented dialogues (Karpiński et al, 2014), as well as for the analysis of cross-modal interactions between interlocutors' hand gestures, gaze shifts and speaking rates (Czoska et al, 2015). Only data provided by adult speakers were used as study material for the preliminary analyses.…”

“…The directions and sizes of these changes are compared between the dialogue participants in order to detect similarities or divergence. The time window size and step size are normally dependent on varying distributions of conversational activities, the duration of the entire dialogue under analysis, as well as the size of the units to be studied (e.g., gesture phrases or phases; Karpiński et al, 2014). A range of factors may influence the distribution of conversational units, their sizes and other properties.…”

“…Following earlier studies in speech alignment by the authors (e.g. Czoska, Klessa, Karpiński, & Nowikow-Jarmołowicz, 2015;Karpiński, Klessa, & Czoska, 2014), this paper analyses gesture units for their alignment in the course of two types of dialogue tasks: a collaborative and a competitive task. All the analyses described in the present paper are based on the Borderland corpus data, collected as part of a project dedicated to the investigation of the paralinguistic features of interpersonal communication in the borderland region of Słubice (Poland) and Frankfurt Oder (Germany) -on the border of languages and cultures (e.g., Karpiński & Klessa, 2018).…”