Refine
Document Type
- Article (6)
- Postprint (6)
- Doctoral Thesis (1)
Language
- English (13)
Is part of the Bibliography
- yes (13)
Keywords
- speech (13) (remove)
Institute
Quantifying neurological disorders from voice is a rapidly growing field of research and holds promise for unobtrusive and large-scale disorder monitoring. The data recording setup and data analysis pipelines are both crucial aspects to effectively obtain relevant information from participants. Therefore, we performed a systematic review to provide a high-level overview of practices across various neurological disorders and highlight emerging trends. PRISMA-based literature searches were conducted through PubMed, Web of Science, and IEEE Xplore to identify publications in which original (i.e., newly recorded) datasets were collected. Disorders of interest were psychiatric as well as neurodegenerative disorders, such as bipolar disorder, depression, and stress, as well as amyotrophic lateral sclerosis amyotrophic lateral sclerosis, Alzheimer's, and Parkinson's disease, and speech impairments (aphasia, dysarthria, and dysphonia). Of the 43 retrieved studies, Parkinson's disease is represented most prominently with 19 discovered datasets. Free speech and read speech tasks are most commonly used across disorders. Besides popular feature extraction toolkits, many studies utilise custom-built feature sets. Correlations of acoustic features with psychiatric and neurodegenerative disorders are presented. In terms of analysis, statistical analysis for significance of individual features is commonly used, as well as predictive modeling approaches, especially with support vector machines and a small number of artificial neural networks. An emerging trend and recommendation for future studies is to collect data in everyday life to facilitate longitudinal data collection and to capture the behavior of participants more naturally. Another emerging trend is to record additional modalities to voice, which can potentially increase analytical performance.
Quantifying neurological disorders from voice is a rapidly growing field of research and holds promise for unobtrusive and large-scale disorder monitoring. The data recording setup and data analysis pipelines are both crucial aspects to effectively obtain relevant information from participants. Therefore, we performed a systematic review to provide a high-level overview of practices across various neurological disorders and highlight emerging trends. PRISMA-based literature searches were conducted through PubMed, Web of Science, and IEEE Xplore to identify publications in which original (i.e., newly recorded) datasets were collected. Disorders of interest were psychiatric as well as neurodegenerative disorders, such as bipolar disorder, depression, and stress, as well as amyotrophic lateral sclerosis amyotrophic lateral sclerosis, Alzheimer's, and Parkinson's disease, and speech impairments (aphasia, dysarthria, and dysphonia). Of the 43 retrieved studies, Parkinson's disease is represented most prominently with 19 discovered datasets. Free speech and read speech tasks are most commonly used across disorders. Besides popular feature extraction toolkits, many studies utilise custom-built feature sets. Correlations of acoustic features with psychiatric and neurodegenerative disorders are presented. In terms of analysis, statistical analysis for significance of individual features is commonly used, as well as predictive modeling approaches, especially with support vector machines and a small number of artificial neural networks. An emerging trend and recommendation for future studies is to collect data in everyday life to facilitate longitudinal data collection and to capture the behavior of participants more naturally. Another emerging trend is to record additional modalities to voice, which can potentially increase analytical performance.
This dissertation focuses on the handling of time in dialogue. Specifically, it investigates how humans bridge time, or “buy time”, when they are expected to convey information that is not yet available to them (e.g. a travel agent searching for a flight in a long list while the customer is on the line, waiting). It also explores the feasibility of modeling such time-bridging behavior in spoken dialogue systems, and it examines
how endowing such systems with more human-like time-bridging capabilities may affect humans’ perception of them.
The relevance of time-bridging in human-human dialogue seems to stem largely from a need to avoid lengthy pauses, as these may cause both confusion and discomfort among the participants of a conversation (Levinson, 1983; Lundholm Fors, 2015). However, this avoidance of prolonged silence is at odds with the incremental nature of speech production in dialogue (Schlangen and Skantze, 2011): Speakers often start to verbalize their contribution before it is fully formulated, and sometimes even before they possess the information they need to provide, which may result in them running out of content mid-turn.
In this work, we elicit conversational data from humans, to learn how they avoid being silent while they search for information to convey to their interlocutor. We identify commonalities in the types of resources employed by different speakers, and we propose a classification scheme. We explore ways of modeling human time-buying behavior computationally, and we evaluate the effect on human listeners of embedding this behavior in a spoken dialogue system.
Our results suggest that a system using conversational speech to bridge time while searching for information to convey (as humans do) can provide a better experience in several respects than one which remains silent for a long period of time. However, not all speech serves this purpose equally: Our experiments also show that a system whose time-buying behavior is more varied (i.e. which exploits several categories from the classification scheme we developed and samples them based on information from human data) can prevent overestimation of waiting time when compared, for example, with a system that repeatedly asks the interlocutor to wait (even if these requests for waiting are phrased differently each time). Finally, this research shows that it is possible to model human time-buying behavior on a relatively small corpus, and that a system using such a model can be preferred by participants over one employing a simpler strategy, such as randomly choosing utterances to produce during the wait —even when the utterances used by both strategies are the same.