TY - JOUR A1 - Sehr, Armin A1 - Barfuss, Hendrik A1 - Hofmann, Christian A1 - Maas, Roland A1 - Kellermann, Walter T1 - Efficient training of acoustic models for reverberation-robust medium-vocabulary automatic speech recognition JF - 2014 4th Joint Workshop on Hands-Free Speech Communication and Microphone Arrays, HSCMA, 12-14 May 2014, Villers-les-Nancy, France N2 - A recently proposed concept for training reverberation-robust acoustic models for automatic speech recognition using pairs of clean and reverberant data is extended from word models to tied-state triphone models in this paper. The key idea of the concept, termed ICEWIND, is to use the clean data for the temporal alignment and the reverberant data for the estimation of the emission densities. Experiments with the 5000-word Wall Street Journal corpus confirm the benefits of ICEWIND with tied-state triphones: While the training time is reduced by more than 90%, the word accuracy is improved at the same time, both for room-specific and multi-style hidden Markov models. Since the acoustic models trained with ICEWIND need less Gaussian components for the emission densities to achieve comparable recognition rates as Baum-Welch acoustic models, ICEWIND also allows for a reduced decoding complexity. KW - Hidden Markov models KW - Training KW - Speech recognition KW - Speech KW - Vectors KW - Accuracy KW - Training data Y1 - 2014 U6 - https://doi.org/10.1109/HSCMA.2014.6843275 SP - 177 EP - 181 PB - IEEE ER - TY - CHAP A1 - Yoshioka, Takuya A1 - Sehr, Armin A1 - Delcroix, Marc A1 - Kinoshita, Keisuke A1 - Maas, Roland A1 - Nakatani, Tomohiro A1 - Kellermann, Walter T1 - Survey on approaches to speech recognition in reverberant environments T2 - Asia-Pacific Signal & Information Processing Association annual summit and conference (APSIPA ASC), 2012 : Hollywood, California, USA, 3 - 6 Dec. 2012 N2 - This paper overviews the state of the art in reverberant speech processing from the speech recognition viewpoint. First, it points out that the key to successful reverberant speech recognition is to account for long-term dependencies between reverberant observations obtained from consecutive time frames. Then, a diversity of approaches that exploit the long-term dependencies in various ways is described, ranging from signal and feature dereverberation to acoustic model compensation tailored to reverberation. A framework for classifying those approaches is presented to highlight similarities and differences between them. KW - Speech recognition KW - Hidden Markov models KW - Reverberation KW - Speech KW - Vectors KW - Speech processing Y1 - 2012 SN - 978-0-6157-0050-2 PB - IEEE ER -