Robert Herms

Improving Transcript-Based Video Retrieval Using Unsupervised Language Model Adaptation

One challenge in automated speech recognition is to determine domain-specific vocabulary like names, brands, technical terms etc. by using generic language models. Especially in broadcast news new names occur frequently. We present an unsupervised method for a language model adaptation, which is used in automated speech recognition with a two-pass decoding strategy to improve spoken document retrieval on broadcast news. After keywords are extracted from each utterance, a web resource is queried to collect utterance-specific adaptation data. This data is used to augment the phonetic dictionary and adapt the basic language model. We evaluated this strategy on a data set of summarized German broadcast news using a basic retrieval setup.

A support framework for automated video and multimedia workflows for production and archive

The management of the massive amount of data in video- and multimedia workflows is a hard and expensive work that requires much personnel and technical resources. Our flexible and scalable open source middleware framework offers solution approaches for the automated handling of the ingest and the workflow by an automated acquisition of all available information. By using an XML format to describe the processes, we provide an easy, fast and well-priced solution without the need for specific human skills.

Prediction of Deception and Sincerity from Speech Using Automatic Phone Recognition-Based Features.

As part of the Interspeech 2016 COMPARE challenge, the two different sub-challenges Deception and Sincerity are addressed. The former refers to the identification of deceptive speech whereas the degree of perceived sincerity of speakers has to be estimated in the latter. In this paper, we investigate the potential of automatic phone recognition-based features for these use case scenarios. The speech transcriptions were used to process the appearing tokens (phoneme, silent pause, filled pause) and the corresponding durations. We designed a high-level feature set including the four groups: vowels, phones, pseudo syllables, and pauses. Additionally, we selected suitable predefined acoustic feature sets and fused them with our introduced features showing a positive effect on the prediction. Moreover, the performance is further boosted by refining these fused features using the ReliefF feature selection method. Experiments show that the final systems outperform the baseline results of both sub-challenges.

Aufbereitung augenmedizinischer Bild-, Patienten- und Diagnosedaten zum Zwecke der Forschung - Ethikrichtlinen und deren praktische Umsetzung.

In der vorliegenden Arbeit stellen die Authoren einen technischen Workflow vor, der darstellt wie in der Praxis gesetzliche Vorgaben im Bezug auf ethische Fragestellungen umgesetzt werden können. Dabei wird auf die rechtlichen Grundlagen auf Bundesebene eingegangen, aber auch auf die Besonderheiten auf Länderebene und der lokalen Umsetzung. Am Fallbeispiel der Kooperation zwischen der Juniorprofessur Media Computing an der TU Chemnitz und der Augenklinik des Klinikum Chemnitz gGmbH zeigen wir dabei welche Vorgaben seitens des Ethikbeauftragten einzuhalten waren hinsichtlich der Anonymisierung von Patientendaten, der Verschlüsselung, dem Transport/Transfer von der Klinik an die Universität, Speicherung und Zugriffsrechte der Daten. Eingegangen wird insbesondere auf unterschiedliche Aspekte in der retrospektiven Forschung mit Patientendaten. Damit soll insbesondere Einsteigern auf dem Gebiet der Forschung mit klinischen Daten ein erster Einblick ermöglicht werden.