12 July: fyi -- speech, France
Index of July 2001 | Index of year: 2001 | Full index
The post-doctoral position described below is available in the speech
synthesis team of FTR&D in Lannion (France). Post-doctoral position in voice
conversion France Télécom R&D focuses its activities in text to speech
synthesis upon a system based on the concatenation of acoustic units. Those
research and development actions has led to the implementation of
operational services such as the mail Itinéris or the reversed phone book
QuiDonc.
In order to develop its offer in speech synthesis and to customize its vocal
services, FTR&D aims to diversify its range of synthesis voices. Moreover,
its current R&D efforts are based around a text to speech system using
variable length units. This technique offers an appreciable improvement in
terms of quality (as compared to diphone-based systems) but increases the
computational load as well as the database size. This latter point is really
crucial because recording a database is always an expensive and tedious
operation. In order to limit this task, a solution consists in developing
voice conversion techniques. This can be accomplished by transforming an
acoustic dictionary. Starting from a small recording of the "target"
speaker, the aim is to modify a "reference" corpus so that the obtained
dictionary seems to have been recorded by the target speaker.
The implementation of such a system can be separated into two distinct
steps. First, a statistical learning is done on the reference corpus as well
as on the target speaker recording. These estimated parameters are then used
to calculate a warping function which can be applied to the reference
dictionary in order to achieve the voice conversion.
In a first stage the voice conversion will be limited to the timbre
modification (spectral envelope). Then it is possible to enhance the system
by incorporating the modification of prosodic parameters (pitch and
duration).
This post-doctoral position is available for one year and will be carried
out in the laboratory "Interaction par la Parole et les Sons" of the FTR&D
center located in Lannion.
Contact :
Olivier Rosec
FTR&D DIH/IPS/VMI
2, avenue Pierre Marzin
22307 Lannion Cedex
Tél : +33 2 96 05 20 67, Fax +33 2 96 05 35 30
e-mail : olivier.rosec@rd.francetelecom.com
Index of July 2001 | Index of year: 2001 | Full index