Abstract
The European project SpeechDat-Car was launched by the European Commission in April 1998 with the purpose of collecting robust speech databases in nine different European languages to support training and testing of robust multilingual speech recognition for in-vehicle applications. After 4 months of extension and a new added language (American English) most of the databases have been recorded and are being processed and validated by SPEX according to the pre-defined SpeechDat-Car standards.
In order to show sufficiency and quality of SpeechDat-Car databases from an end user point of view, some field trials have been planned in collaboration with speech technology providers. The purpose of these trials is to compare performances of a 'basic' speech recogniser with one especially trained with SpeechDat-Car data for different recognition technologies and for different languages.
This paper describes the field trials, that include the recording of additional test databases in different platforms, the experiments performed and the final results and conclusions.
Finally, some plans for exploitation of SpeechDat-Car results are summarised.