A Corpus of Read and Spontaneous Upper Saxon German Speech for ASR Evaluation

A Corpus of Read and Spontaneous Upper Saxon German Speech for ASR Evaluation Robert Herms1, Laura Seelig1, Stefanie M¨unch2, and Maximilian Eibl1 1Chair Media Informatics, Technische Universit¨at Chemnitz, Germany 2FIITS, University of Applied Sciences Dresden, Germany {robert.herms, maximilian.eibl}@cs.tu-chemnitz.de Abstract In this Paper we present a corpus named SXUCorpus which contains read and spontaneous speech of the Upper Saxon German dialect. The data has been collected from eight archives of local television stations located in the Free State of Saxony. The recordings include broadcasted topics of news, economy, weather, sport, and documentation from the years 1992 to 1996 and have been manually transcribed and labeled. In the paper, we report the methodology of collecting and processing analog audiovisual material, constructing the corpus and describe the properties of the data. In its current version, the corpus is available to the scientific community and is designed for automatic speech recognition (ASR) evaluation with a development set and a test set. We performed ASR experiments with the open-source framework sphinx-4 including a configuration for Standard German on the dataset. Additionally, we show the influence of acoustic model and language model adaptation by the utilization of the development set. Keywords: speech corpus, speech recognition, dialect 1. Introduction The collection of regional speech patterns and its prepara- tion is the basis for many fields in linguistic research as well as the development and evaluation of speech technol- ogy applications. The dialect Upper Saxon is attributed to the East Central German dialect groups. It is mainly spoken in the Free State of Saxony and adjacent areas as the east- ern part of Thuringia and the south-eastern Saxony-Anhalt. Upper Saxon has major phonological, morphological and lexical deviations from Standard German and other regi- olects (Weise and others, 2013). Furthermore, there are re- gional variants within Saxony (Schaufuß, 2015) which arise particularly in spontaneous speech. Some recordings of the dialect are already available by different German speech corpora. For instance, the cor- pus RVG 1 (Burger and Schiel, 1998) includes recordings from all over Germany. It contains read and spontaneous speech that were orthographically and phonetically tran- scribed. The corpus of FOLK (Schmidt, 2014) contains over 100 hours of different verbal interaction types from different regions including Upper Saxon. In other works, Upper Saxon speech has been collected in connection with specific research activities, such as the intonation of a di- alect (K¨ugler, 2003), or standard-dialect variation (Schau- fuß, 2015). The project “Pilotprojekt zur Digitalisierung der Senderar- chive s¨achsischer Lokalfernsehsender” in cooperation with the SLM (S¨achsische Landesanstalt f¨ur privaten Rundfunk und neue Medien) tries to support local television stations by developing solutions for archiving analog audiovisual material from the early nineties. In this context, we built a comprehensive workflow including the digitization and an- notation in order to provide a holistic approach. As part of the workflow, the automatic speech recognition (ASR) allows the retrieval of spoken language. However, the chal- lenge is to recognize dialect speech. In this Paper we present the first version of the SXUCorpus which …

Content truncated to 500 words