Simulating Intonational Varieties of Swedish
This project represents collaboration between Lund University (Gösta Bruce and Susanne Schötz) and KTH (Björn Granström, Jonas Beskow, Joakim Gustafson and Laura Enflo) The primary goal of the research project is to produce more precise and thorough knowledge about phrase and utterance prosody in some major regional varieties of Swedish. A fundamental idea in the project work is to take advantage of speech synthesis in different forms in order to be able to reach our goals.
The primary goal of the research project is to produce more precise and thorough knowledge about phrase and utterance prosody in some major regional varieties of Swedish. In this research effort the Swedish prosody model plays a prominent role. An accompanying goal is thus to develop our model both theoretically and experimentally. A fundamental idea in the project work is to take advantage of speech synthesis in different forms in order to be able to reach our goals.
Our project will be based primarily on already recorded speech material. The main project work will consist of the analysis and description of this material regarding prosody as well as the simulation of prosodic patterns by means of speech synthesis. In our analysis and synthesis work the idea is to focus on some major intonational types. These types are supposed to be geographically distributed: South, Göta, Svea, Gotland, Dala, North, and Finland Swedish.
The significance of our project work will be both within basic research and within speech technology applications. Thus our project is expected to contribute to deepening our understanding of prosodic structuring of spoken language generally, and more specifically of dialectal variation of Swedish. It is likely to contribute to generating new knowledge about prosody to be used in speech synthesis systems. The project has the potential of becoming a pioneer enterprise generally within the field of synthetic speech with its aim to be able to simulate different accents of a language.
In the project a hybrid system was developed where an HMM synthesizer trained on a speaker with a certain dialect is used to generate tracks for pitch, duration and phonetic content. Prosodic rules for the same dialect are then applied to get a rule based pitch curve that is merged with the data-driven pitch curve before rendering the waveform.
Group: Speech Communication and Technology
Björn Granström (Project leader)
Gösta Bruce (Project leader)
Duration: 2007 - 2010
Keywords: prosody, dialect, intonation, regional, synthesis
Alternative Measures of Phonation: Collision Threshold Pressure and Electroglottographic Spectral Tilt. Extra: Perception of Swedish Accents. Licentiate dissertation, School of Computer Science and Communication, KTH. [pdf] (2010).
Speech Prosody 2010. Chicago, USA. [pdf] (2010). Simulating Intonation in Regional Varieties of Swedish. In
Fonetik 2010. Lund, Sweden. [pdf] (2010). Simulating Intonation in Regional Varieties of Swedish. In
Proceedings of Fonetik 2009. (2009). Experiments with Synthesis of Swedish Dialects. In
Proceedings of Fonetik 2008. (2008). Human Recognition of Swedish Dialects. In
Proceedings of Interspeech 2008. [pdf] (2008). Recognizing and Modelling Regional Varieties of Swedish. In
Proceedings of Perception and Interactive Technologies for Speech-Based Systems (PIT 2008) (pp. 293-296). Berlin/Heidelberg: Springer. [abstract] [pdf] (2008). expros: a toolkit for exploratory experimentation with prosody in customized diphone voices. In
Proceedings of FONETIK 2008 (pp. 17-20). Gothenburg, Sweden. [abstract] [pdf] (2008). EXPROS: Tools for exploratory experimentation with prosody. In