Multi-Staged Cross-Lingual Acoustic Model Adaption for Robust Speech Recognition in Real-World Applications - A Case Study on German Oral History Interviews

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6354–6362 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6354 Multi-Staged Cross-Lingual Acoustic Model Adaption for Robust Speech Recognition in Real-World Applications — A Case Study on German Oral History Interviews Michael Gref,1,2 Oliver Walter,1 Christoph Schmidt,1 Sven Behnke,1,3 Joachim K¨ohler1 1Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS), Germany 2Institute for Pattern Recognition (iPattern), Niederrhein Univ. of Applied Sciences, Germany 3Autonomous Intelligent Systems (AIS), Computer Science Institute VI, Univ. of Bonn, Germany {michael.gref, oliver.walter, christoph.andreas.schmidt, sven.behnke, joachim.koehler}@iais.fraunhofer.de Abstract 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. Keywords: acoustic modeling, acoustic model adaption, cross-lingual, digital humanities, oral history, speech recognition, transfer- learning, under-resourced speech recognition 1. Introduction Current automatic speech recognition (ASR) systems show remarkable performance on tasks where large amount of annotated and time-aligned speech data is used for training. Other tasks, however, where only little or no data is present—such as under-resourced or zero resource languages—are still a challenge for modern systems. Nev- ertheless, also within one language, the recognition perfor- mance drastically degrades if off-the-shelf systems are ap- plied to different domains in terms of recording conditions and speech characteristics, such as noisy and reverberated recordings, spontaneous, fast speech, accents, unclear pro- nunciations and age- and health related changes in the ways of speaking—to name but a few. In this work, we propose and investigate an approach that performs a robust acoustic model adaption to a target domain in a cross-lingual, multi- staged manner. The proposed approach utilizes large-scale training data from rich-resourced domains in both the same and other languages. A lot of investigations in the field of speech recognition are based on well-known, publicly available benchmark data sets like Switchboard, Wall Street Journal and AMI to eval- uate proposed systems and approaches. Usually, such data sets cover well-defined, restricted challenges and are of good use to study this specific challenge. However, they are only of limited use for replicating the conditions of chal- …

Content truncated to 500 words