New telephone speech databases for French: a children database and an optimized adult corpus

This paper presents the results of the NEOLOGOS project: a children database and an optimized adult database for the French language. A new approach was adopted for the collection of the adult database in order to enable the development of new algorithms in the field of speech processing (study of speaker characteristics, speakers similarity, speaker selection algorithms,...) The objective here was to define and to carry out a new methodology for collecting significant quantities of speaker dependent data, for a significant number of speakers, as was done for several databases oriented towards speaker verification, but with the additional constraint of maximising the coverage of the space of all speakers. The children database is made of 1,000 sessions recorded by children between 7 and 16 years old. Both speach databases are SpeehDat compliant meaning that they can be easily used for research and development in the field of speech technology. 1