Where academic tradition
meets the exciting future

You are here: TUCS > PUBLICATIONS > Publication Search > Word Model-Determined Segmenta...

Word Model-Determined Segmental Duration in Finnish Speech Synthesis and its Effect on Naturalness

Jussi Hakokari, Tuomo Saarni, Mikko Jalonen, Olli Aaltonen, Jouni Isoaho, Tapio Salakoski, Word Model-Determined Segmental Duration in Finnish Speech Synthesis and its Effect on Naturalness. In: Proceedings of the Second Baltic Conference on Human Language Technologies (HLT 2005), 2005.

Abstract:

The 50-year-old concept of formant synthesis was under much scientific scrutiny until it became
apparent that naturalness was hard to attain using rule-based synthesis methods. Formant
synthesis was essentially an intelligible approximation, rather than imitation, of human speech.
Our current research is a revisitation to the rule-based formant synthesis. At the moment, due to
advances in computer technology, we are in a better disposition to develop naturalness in rulebased
text-to-speech systems. While research teams in various countries have made similar
efforts, our project is unique in Finland and, more importantly, as regarding the Finnish language.
The duration of individual phones is important to naturalness in Finnish. We have extracted data
from an extensive single-speaker Finnish speech corpus and created word models to prescribe
each phone a duration depending on its position within a word. In this paper, we will describe our
approach, present the preliminary results of listening tests and discuss the potential of word
models in improving naturalness in text-to-speech systems.

Files:

Abstract in PDF-format

BibTeX entry:

@INPROCEEDINGS{inpHaSaJaAaIsSa05a,
  title = {Word Model-Determined Segmental Duration in Finnish Speech Synthesis and its Effect on Naturalness},
  booktitle = {Proceedings of the Second Baltic Conference on Human Language Technologies (HLT 2005)},
  author = {Hakokari, Jussi and Saarni, Tuomo and Jalonen, Mikko and Aaltonen, Olli and Isoaho, Jouni and Salakoski, Tapio},
  year = {2005},
}

Belongs to TUCS Research Unit(s): Turku BioNLP Group, Communication Systems (ComSys)

Edit publication