IARPA BABEL Program

The IARPA BABEL Program ran from 2012-2016. The goal of the program was to improve keyword spotting in low-resource languages with limited training data. While word error rates could be as high as 50% for some of the languages, we were still able to detect keywords with a high accuracy.

Related Publications

• “Analysis of keyword spotting performance across IARPA babel languages”, William Hartmann, Damianos Karakos, Roger Hsiao, Le Zhang, Alumäe, Stavros Tsakalidis, Richard Schwartz, in Proceedings of IEEE ICASSP, pp. 5765-5769, 2017. [publication] [poster]

• “The 2016 BBN Georgian telephone speech keyword spotting system”, Tanel Alumäe, Damianos Karaks, William Hartmann, Roger Hsiao, Le Zhang, Long Nguyen, Stavros Tsakalidis, Richard Schwartz, in Proceedings of IEEE ICASSP, pp. 5755-5759, 2017. [publication]

• “Alternative networks for monolingual bottleneck features”, William Hartmann, Roger Hsiao, Stavros Tsakalidis, in Proceedings of IEEE ICASSP, pp. 5290-5294, 2017. [publication]

• “Two-Stage Data Augmentation for Low-Resourced Speech Recognition”, William Hartmann, Tim Ng, Roger Hsiao, Stavros Tsakalidis, Richard Schwartz, in Proceedings of Interspeech, pp. 2378-2382, 2016. [publication] [slides]

• “Comparison of Multiple System Combination Techniques for Keyword Spotting”, William Hartmann, Le Zhang, Kerri Barnes, Roger Hsiao, Stavros Tsakalidis, Richard Schwartz, in Proceedings of Interspeech, pp. 1913-1917, 2016. [publication] [poster]

• “Enhancing low resource keyword spotting with automatically retrieved web documents”, Le Zhang, Damianos Karakos, William Hartmann, Roger Hsiao, Richard Schwartz, Stavros Tsakalidis, in Proceedings of Interspeech, pp. 839-843, 2015. [publication]

• “Exploring minimal pronunciation modeling for low resource languages”, Marelie Davel, Etienne Barnard, Charl van Heerden, William Hartmann, Damianos Karakos, Richard Schwartz, Stavros Tsakalidis, in Proceedings of Interspeech, pp. 538-542, 2015. [publication]

• “Cross-word sub-word units for low-resource keyword spotting”, William Hartmann, Lori Lamel, Jean-Luc Gauvain, in Proceedings of SLTU, pp. 112–117, 2014. [publication] [post]

• “Comparing decoding strategies for subword-based keyword spotting in low-resourced languages”, William Hartmann, Viet-Bac Le, Abdel Messaoudi, Lori Lamel, Jean-Luc Gauvain, in Proceedings of Interspeech, pp. 2764-2768, 2014. [publication] [post]

• “Developing STT and KWS systems using limited language resources”, Viet-Bac Le, Lori Lamel, Abdel Messaoudi, William Hartmann, Jean-Luc Gauvain, Cécile Woehrling, Julien Despres, Anindya Roy, in Proceedings of Interspeech, pp. 2484-2488, 2014. [publication]