The BAS Speech Data Repository

The BAS Speech Data Repository Uwe D. Reichel, Florian Schiel, Thomas Kisler, Christoph Draxler, Nina P¨orner Research Institute for Linguistics, Hungarian Academy of Sciences, Institute of Phonetics and Speech Processing, University of Munich Bencz´ur u. 33, 1068 Budapest, Hungary, Schellingstr. 3, 80799 Munich, Germany uwe.reichel@nytud.mta.hu, {schiel,kisler,draxler,poerner}@bas.uni-muenchen.de Abstract The BAS CLARIN speech data repository is introduced. At the current state it comprises 31 pre-dominantly German corpora of spoken language. It is compliant to the CLARIN-D as well as the OLAC requirements. This enables its embedding into several infrastructures. We give an overview over its structure, its implementation as well as the corpora it contains. Keywords: Data repository, spoken language corpora, CLARIN-D 1. Introduction The BAS CLARIN speech data repository (BAS, 2016a; Reichel, 2013) is maintained by the Bavarian Archive for Speech Signals (BAS, 2016c) in the context of the CLARIN-D project (CLARIN, 2016b). It is located at the Institute of Phonetics and Speech Processing, Ludwig- Maximilian-University in Munich, Germany. At its current state (March 2016) it comprises 31 pre-dominantly German corpora of spoken language summing up to 2.62 TB of sig- nal and annotation files, and 16.6 GB of metadata. 29 of these corpora are freely available for academic research. The repository is compliant to the CLARIN-D as well as the OLAC requirements, so that it can be harvested by sev- eral infrastructures such as the Virtual language observatory (VLO, 2015; Goosen and Eckart, 2014) and the Open Lan- guage Archives Community (OLAC, 2011). Figures 1 and 2 show the repository’s and a corpus’ landing page, respec- tively. This paper describes the structure of this repository, its main features, and gives an overview over the provided cor- pora. Figure 1: The repository’s landing page. 2. Repository structure and main features 2.1. Structure The repository is based on a file system and is hierarchically structured into corpus, recording session and primary data items. Each corpus contains one or more recording ses- sions, and each session comprises primary data, i.e. signal and annotation files. The repository is divided into a freely accessible and a protected part. The protected part contains the primary resources, whereas the metadata is freely ac- cessible. In compliance with the CLARIN-D requirements the BAS repository adheres to standardized file formats, provides metadata, supports persistent data storage and versioning, and requires user authentication for its protected part. 2.2. Metadata Each corpus and each recording session is described by a CMDI metadata record (CLARIN, 2016d) that is dynami- cally rendered to a landing page for this item to be accessed by the users. Next to CMDI also Dublin Core and OLAC format (DC, 2016; OLAC, 2011) are supported. The meta- data can be harvested via an OAI-PMH endpoint (BAS, 2016b). 2.3. Persistence The BAS Repository supports a persistent storage of the contained data: each version of a repository item is per- manently stored without changes. Primary data is stored together with its MD5 checksum (Rivest, 1992) so that con- sistency can be regularly checked. Furthermore, each version of a corpus and …

Content truncated to 500 words