RVG 1 - A Database for Regional Variants of Contemporary German
Susanne Burger, Florian Schiel · 2007
Regional speaker variability is a major problem in today's stateof -the-art speech recognition systems. Therefore, a major point in the creation of speech resources is the regional coverage of data within one language. At the beginning of 1996 we started to collect data for the RVG1 (Regional Variants of German) corpus. This project was established in cooperation between the American telephone company AT&T, Lucent Technologies and the Bavarian Archive for Speech Signals -- BAS, Munich (Schiel, 1997). It can be seen as a first small database of regionally covered recordings of German representing the most common dialectal regions or at least all those regions which could be important for categorising regional variants into broader classes. RVG1 contains read numbers, phonetically rich sentences and computer commands as well as spontaneous speech. The speech signals are recorded in parallel in four different sound qualities. This paper documents the tasks, the criteria for speaker select...