Adaptation of Automatic Speech Recognition systems to the needs of Austrian German

State-of-the-art automatic speech recognition (ASR) systems usually do not distinguish between Austrian Standard German (ASG) and German Standard German (GSG). Though both called German, it is obvious that differences between GSG and ASG are not negligible [2]. Unsurprisingly, recognition results of such ASR systems that were trained on GSG are worse than necessary when dealing with ASG that is taken from the GRASS Corpus [5]. In this study, it is shown that already small adjustments of GSG clearly improve recognition results for read speech (RS) of ASG. These adjustments consist of a set of rules that account for the typical differences between ASG and GSG, yielding a better representation of ASG pronunciation [4]. When it comes to conversational speech (CS), additional aspects need to be considered as many rules that apply for RS are changed, replaced or disregarded by speakers having a conversation. These changes are usually dependent on several factors such as the speakers' relation, the conversational context etc. and vary from speaker to speaker [3]. To overcome this broad range of "exceptions" – compared to RS – there is need to account for both standard rules and reduction or transformation rules as well as dialect words [1].

References
[1] Sylvia Moosmüller. "Die österreichische Variante der Standardsprache". In: Beiträge
zur deutschen Standardaussprache. Hanau/Halle, Werner Dausien Verlag (1996),
pp. 204–213.
[2] Sylvia Moosmüller. "Hochsprache und Dialekt in Österreich". Vol. 1. Böhlau, 1991.
[3] Rudolf Muhr. "The pragmatics of a pluricentric language: A comparison between
Austrian German and German German". In: Pragmatics and Beyond New Series 178 (2008), p. 211.
[4] Barbara Schuppler, Martine Adda-Decker, and Juan A Morales-Cordovilla. "Pronunciation
variation in read and conversational Austrian German". In: Fifteenth
Annual Conference of the International Speech Communication Association. 2014.
[5] Barbara Schuppler et al. "GRASS: the Graz corpus of Read And Spontaneous
Speech." In: LREC. 2014, pp. 1465–1470.