New results on automatic speech recognition in extremely reverberant environments

Hofmann C, Sehr A, Sahoo SK, Maas R, Kellermann W (2013)

Publication Language: English

Publication Type: Conference contribution

Publication year: 2013

City/Town: Merano, Italy

Pages Range: 2059-2062

Event location: Merano IT

ISBN: 978-3-9392-9605-8

Abstract

Today’s life offers a plethora of devices equipped with an acoustic human-machine interface. These interfaces, providing comfortable access to smartphones, navigation devices, entertainment systems, and many more, require a high recognition accuracy of the underlying Automatic Speech Recognition (ASR) system. Due to the interfering signals recorded by distant microphones, reliable ASR is in general hard to achieve in real-world acoustic environments. While outdoor applications suffer mainly from noise, indoor applications are additionally impaired by reverberation. The reverberation of typical office or lecture-room environments already renders small-vocabulary recognizers trained on clean speech unusable. Nevertheless, ASR technology would also be highly desirable in very large and reverberant environments, such as large entrance halls or exhibition rooms in museums, e.g., for interaction with a voice-controlled information retrieval system. This contribution explains why such environments do not necessarily preclude the use of ASR systems and provides experimental verification of ASR in an extremely reverberant environment along with a dedicated and efficient ASR training method known as ICEWIND.

Authors with CRIS profile

Christian Hofmann Professur für Signalverarbeitung Armin Sehr Professur für Signalverarbeitung Roland Maas Lehrstuhl für Multimediakommunikation und Signalverarbeitung (LMS) Walter Kellermann Professur für Signalverarbeitung

How to cite

APA:

Hofmann, C., Sehr, A., Sahoo, S.K., Maas, R., & Kellermann, W. (2013). New results on automatic speech recognition in extremely reverberant environments. In Proceedings of the AIA-DAGA Conference on Acoustics (pp. 2059-2062). Merano, IT: Merano, Italy.

MLA:

Hofmann, Christian, et al. "New results on automatic speech recognition in extremely reverberant environments." Proceedings of the AIA-DAGA Conference on Acoustics, Merano Merano, Italy, 2013. 2059-2062.

BibTeX: Download