Speaker Localization for Microphone ArrayBased ASR The Effects of Accuracy on Overlapping.pdfVIP

Speaker Localization for Microphone Array-Based ASR: The Effects of Accuracy on Overlapping Speech Hari Krishna Maganti 12 and Daniel Gatica-Perez13 1 IDIAP Research Institute, Martigny, Switzerland 2 University of Ulm, Ulm, Germany 3 Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland Rue du Simplon 4, Martigny, Switzerland {hari,gatica}@idiap.ch ABSTRACT computer interaction applications in many conversational Accurate speaker location is essential for optimal perfor- settings, including meetings. Meetings represent an effec- tive mode of information exchange among humans. Auto- mance of distant speech acquisition systems using micro- phone array techniques. However, to the best of our knowl- matic analysis of meetings have a potential of conveying edge, no comprehensive studies on the degradation of au- detailed knowledge about human activities which can also tomatic speech recognition (ASR) as a function of speaker be archived for later retrieval and review [4]. location accuracy in a multi-party scenario exist. In this pa- In the context of automatic analysis of meetings, robust per, we describe a framework for evaluation of the effects of localization and tracking of active speakers is of fundamen- speaker location errors on a microphone array-based ASR tal importance, particularly for enhancemen