Data-Driven Pronunciation Modeling of Swiss German Dialectal Speech for Automatic Speech Recognition

Data-Driven Pronunciation Modeling of Swiss German Dialectal Speech for Automatic Speech Recognition Michael Stadtschnitzer, Christoph Schmidt Fraunhofer IAIS, Schloss Birlinghoven, 53757 Sankt Augustin {michael.stadtschnitzer, christoph.andreas.schmidt}@iais.fraunhofer.de Abstract Automatic speech recognition is a requested technique in many fields like automatic subtitling, dialogue systems and information retrieval systems. The training of an automatic speech recognition system is usually straight forward given a large annotated speech corpus for acoustic modeling, a phonetic lexicon, and a text corpus for the training of a language model. However, in some use cases these resources are not available. In this work, we discuss the training of a Swiss German speech recognition system. The only resources that are available is a small size audio corpus, containing the utterances of highly dialectical Swiss German speakers, annotated with a standard German transcription. The desired output of the speech recognizer is again standard German, since there is no other official or standardized way to write Swiss German. We explore strategies to cope with the mismatch between the dialectal pronunciation and the standard German annotation. A Swiss German speech recognizer is trained by adapting a standard German model, based on a Swiss German grapheme-to-phoneme conversion model, which was learned in a data-driven manner. Also, Swiss German speech recognition systems are created, with the pronunciation based on graphemes, standard German pronun- ciation and with a data-driven Swiss German pronunciation model. The results of the experiments are promising for this challenging task. Keywords: Robust Speech Recognition, Swiss German, Dialectal Speech, Data-Driven Pronunciation Modeling 1. Introduction Switzerland has four national languages: German/Swiss German (63%), French (22.7%), Italian (8.1%), Romansh (0.5%); the numbers in brackets are the percentages of the population speaking them1. As can be derived from Figure 1, French is spoken in the west, Italian is spoken primar- ily in Ticino, Val Bregaglia and Val Pschiavo and Romansh speakers are distributed over Graub¨unden. Swiss German, which is primarily spoken in the center and east of Switzer- land, is highly dialectical. Typically, speakers speak a di- alect representative of the region. To be understood by vis- itors, Swiss German speakers switch to standard German (Garner et al., 2014). Swiss German and its dialectal vari- ants do not have a standard written form, instead the stan- dard written form is standard German. Although regional Swiss German dialects are manifold and their differences can be very subtle, there is a common Swiss German speak- ing style which is used in Swiss broadcasts (e.g. weather re- ports of Schweizer Radio und Fernsehen2 (SRF)), and that is understood well by the vast majority of the Swiss Ger- man speaking people. Broadcast companies are typically interested in the automatic transcription of speech data in- cluding (live) subtitling and automatic speech recognition (ASR). In the case of Swiss German, the desired output of the ASR system is standard German, as there is no stan- dardized written form of Swiss German dialect. Due to the mismatch between the dialectal pronunciation of Swiss German and the written form of standard German, these system often …

Content truncated to 500 words