We introduce the NU-MBROLA project as an extension of the MBROLA project for designing large NU-MBROLA databases exhibiting the same properties as MBROLA diphone databases, without the restriction of being composed of diphones. The accompanying NU-MBROLA synthesizer implements the MBROLA algorithm on speech segments only defined by their starting and ending points in natural speech files. It is distributed with the same terms and conditions as in the MBROLA project. The terms and conditions for the creation of NU-MBROLA database are slightly different from those related to the creation of MBROLA databases, mainly in that no speech segmentation is required, and in that we leave it to providers to distribute their NU-MBROLA databases.

Speech synthesis is an automatic encoding process carried out by machine through which symbols conveying linguistic information are converted into an acoustic waveform. In the past decade or so, a recent trend toward a non-parametric, corpus-based approach has focused on using real human speech as source material for producing novel natural-sounding speech. This work proposes a communication-theoretic formulation in which unit selection is a noisy channel through which an input sequence of symbols passes and an output sequence, possibly corrupted due to the coverage limits of the corpus, emerges. The penalty of approximation is quantified by substitution and concatenation costs which grade what unit contexts are interchangeable and where concatenations are not perceivable. These costs are semi-automatically derived from data and are found to agree with acoustic-phonetic knowledge.

This paper describes the Norwegian speech database FonDat1 designed for development and assessment of Norwegian unit selection speech synthesis. The quality of unit selection speech synthesis systems depends highly on the database used. The database should contain sufficient phonemic and prosodic coverage. High quality unit selection synthesis also requires that the database is annotated with accurate information about identity and position of the units. Traditionally this involves much manual work, either by hand labeling the entire database or by correcting automatic annotations. We are working on methods for a complete automation of the annotation process. To validate these methods a realistic unit selection synthesis database is needed. In addition to serve as a testbed for annotation tools and synthesis experiments, the process of producing the database using automatic methods is in itself an important result. FonDat1 contains studio recordings of approximately 2000 sentences read by two professional speakers, one male and one female. 10 % of the database is manually annotated. 1.

A new method of speech synthesis, which combines earlier work on data-driven formant synthesis with improved data extraction and concatenation of recorded voiceless segments, has been developed and implemented as a TTS system. A listening test has been carried out, which has shown that this hybrid synthesis significantly raises the perception of naturalness in the synthesized speech compared to rule only formant synthesis.

Abstract. The increasing capabilities of experimental household robot platforms require more and more sophisticated methods of interaction. While there are many developments in all directions of Human-Machine-Interaction, the integration and combination of several modalities into one robot system require some effort. To ease the development of applications supporting several types of interaction, Fraunhofer IPA has developed a framework named “Go”. Within this framework we have integrated different kinds of interaction methods into one robot platform “Care-O-bot 3”, a mobile service robot for accomplishing daily tasks. This framework and its interaction methods are presented here. 1

Abstract not found

ISBN 91-7265-901-7 printed version ISBN 91-7265-902-5 web version 2004-05-14