While recent automatic speech recognition systems achieve remarkable performance when large amounts of adequate, high quality annotated speech data is used for training, the same systems often only achieve an unsatisfactory result for tasks in domains that greatly deviate from the conditions represented by the training data. For many real-world applications, there is a lack of sufficient data that can be directly used for training robust speech recognition systems. To address this issue, we propose and investigate an approach that performs a robust acoustic model adaption to a target domain in a cross-lingual, multi-staged manner. Our approach enables the exploitation of large-scale training data from other domains in both the same and other languages. We evaluate our approach using the challenging task of German oral history interviews, where we achieve a relative reduction of the word error rate by more than 30% compared to a model trained from scratch only on the target domain, and 6-7% relative compared to a model trained robustly on 1000 hours of same-language out-of-domain training data.

The newest generation of speech technology caused a huge increase of audio-visual data nowadays being enhanced with orthographic transcripts such as in automatic subtitling in online platforms. Research data centers and archives contain a range of new and historical data, which are currently only partially transcribed and therefore only partially accessible for systematic querying. Automatic Speech Recognition (ASR) is one option of making that data accessible. This paper tests the usability of a state-of-the-art ASR-System on a historical (from the 1960s), but regionally balanced corpus of spoken German, and a relatively new corpus (from 2012) recorded in a narrow area. We observed a regional bias of the ASR-System with higher recognition scores for the north of Germany vs. lower scores for the south. A detailed analysis of the narrow region data revealed – despite relatively high ASR-confidence – some specific word errors due to a lack of regional adaptation. These findings need to be considered in decisions on further data processing and the curation of corpora, e.g. correcting transcripts or transcribing from scratch. Such geography-dependent analyses can also have the potential for ASR-development to make targeted data selection for training/adaptation and to increase the sensitivity towards varieties of pluricentric languages.