An Automatic Prosody Tagger for Spontaneous Speech

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 377–386, Osaka, Japan, December 11-17 2016. An Automatic Prosody Tagger for Spontaneous Speech M´onica Dom´ınguez Universitat Pompeu Fabra C. Roc Boronat, 138 08018, Barcelona, Spain monica.dominguez@upf.edu Mireia Farr´us Universitat Pompeu Fabra C. Roc Boronat, 138 08018, Barcelona, Spain mireia.farrus@upf.edu Leo Wanner ICREA & UPF C. Roc Boronat, 138 08018, Barcelona, Spain leo.wanner@upf.edu Abstract Speech prosody is known to be central in advanced communication technologies. However, despite the advances of theoretical studies in speech prosody, so far, no large scale prosody annotated resources that would facilitate empirical research and the development of empirical computational approaches are available. This is to a large extent due to the fact that current com- mon prosody annotation conventions offer a descriptive framework of intonation contours and phrasing based on labels. This makes it difficult to reach a satisfactory inter-annotator agree- ment during the annotation of gold standard annotations and, subsequently, to create consistent large scale annotations. To address this problem, we present an annotation schema for promi- nence and boundary labeling of prosodic phrases based upon acoustic parameters and a tagger for prosody annotation at the prosodic phrase level. Evaluation proves that inter-annotator agreement reaches satisfactory values, from 0.60 to 0.80 Cohen’s kappa, while the prosody tagger achieves acceptable recall and f-measure figures for five spontaneous samples used in the evaluation of monologue and dialogue formats in English and Spanish. The work presented in this paper is a first step towards a semi-automatic acquisition of large corpora for empirical prosodic analysis. 1 Introduction Speech prosody is known to be central in advanced communication technologies. It is decisive in struc- turing the message, stressing parts of the message that the interlocutor considers important, and reveal- ing information about the interlocutor’s attitude and affection state (Nooteboom, 1997; Wennerstrom, 2001). However, despite the advances of theoretical studies in speech prosody, so far, no sufficiently large, well-annotated prosody material has been created to support empirical studies and drive the re- search on empirical techniques for analysis and generation of prosodic cues, especially for application in human-computer interaction technologies. Common annotation conventions, such as the ToBI conven- tion (Beckman et al., 2005), provide a descriptive framework of intonation contours and phrasing based upon labels that are language-dependent and rather subjective, which makes it difficult to reach a satis- factory inter-annotator agreement for creating gold standard annotations to train and evaluate algorithms. It is, therefore, not surprising that empirical research is still based upon rather small laboratory ex- periments. A further consequence of the lack of sound universal prosody annotation conventions is that current methodologies applied to speech prosody segmentation are still based upon textual and lin- guistic units (usually words or syntax) rather than on acoustic and phonological units (prosodic phrases and prosodic words). These limitations become an insurmountable barrier for technologies that aim at grasping prosodic cues in spontaneous speech, where many complex prosodic, linguistic and affective phenomena occur (hesitations, incoherent discourse structure, false starts, …

Content truncated to 500 words