Where academic tradition
meets the exciting future

The Role of Duration in Finnish Rule-Based TTS

Tuomo Saarni, Jussi Hakokari, Tapio Salakoski, Jouni Isoaho, Olli Aaltonen, The Role of Duration in Finnish Rule-Based TTS. In: Speech Analysis, Synthesis and Recognition, Applications of Phonetics, 2005.

Abstract:

We are developing a rule-based Finnish-language TTS system. Our
primary concern is to find ways to increase naturalness in the synthesis.
Our approach is to observe tendencies in natural language through
acoustic analysis and data mining, and to implement our findings into the
synthesizer. We have concentrated on modeling duration, which is an
essential part of Finnish prosody. The language exhibits contrasting
phonemic lengths and the durations of individual phones are highly
sensitive to their position within a word. We have developed a duration
model (“word models”) based on how the syllabic structure of a word
correlates with segmental durations in a natural speech corpus. We have
implemented and automatized the word models, and studied through
listening tests whether they improve naturalness in the synthesis. We
compared the word model–determined segmental durations with with
fixed ones. The result was ambiguous: the word models appear to
improve naturalness in longer speech stimuli, but not in the shorter ones.

Files:

Abstract in PDF-format

BibTeX entry:

@INPROCEEDINGS{inpSaHaSaIsAa05a,
  title = {The Role of Duration in Finnish Rule-Based TTS},
  booktitle = {Speech Analysis, Synthesis and Recognition, Applications of Phonetics},
  author = {Saarni, Tuomo and Hakokari, Jussi and Salakoski, Tapio and Isoaho, Jouni and Aaltonen, Olli},
  year = {2005},
}

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

Edit publication