Showing 1–2 of 2 results for author: Kang, T G

Adaptable Multi-Domain Language Model for Transformer ASR

Authors: Taewoo Lee, Min-Joong Lee, Tae Gyoon Kang, Seokyeoung Jung, Minseok Kwon, Yeona Hong, Jungin Lee, Kyoung-Gu Woo, Ho-Gyeong Kim, Jiseung Jeong, Jihyun Lee, Hosik Lee, Young Sang Choi

Abstract: We propose an adapter based multi-domain Transformer based language model (LM) for Transformer ASR. The model consists of a big size common LM and small size adapters. The model can perform multi-domain adaptation with only the small size adapters and its related layers. The proposed model can reuse the full fine-tuned LM which is fine-tuned using all layers of an original model. The proposed LM c… ▽ More We propose an adapter based multi-domain Transformer based language model (LM) for Transformer ASR. The model consists of a big size common LM and small size adapters. The model can perform multi-domain adaptation with only the small size adapters and its related layers. The proposed model can reuse the full fine-tuned LM which is fine-tuned using all layers of an original model. The proposed LM can be expanded to new domains by adding about 2% of parameters for a first domain and 13% parameters for after second domain. The proposed model is also effective in reducing the model maintenance cost because it is possible to omit the costly and time-consuming common LM pre-training process. Using proposed adapter based approach, we observed that a general LM with adapter can outperform a dedicated music domain LM in terms of word error rate (WER). △ Less

Submitted 10 February, 2021; v1 submitted 14 August, 2020; originally announced August 2020.

Comments: This paper is accepted for presentation at IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2021

Performance of the CREAM calorimeter in accelerator beam test

Authors: Y. S. Yoon, H. S. Ahn, M. G. Bagliesi, G. Bigongiari, O. Ganel, J. H. Han, H. J. Hyun, J. A. Jeon, T. G. Kang, H. J. Kim, K. C. Kim, J. K. Lee, M. H. Lee, L. Lutz, P. Maestro, A. Malinine, P. S. Marrocchesi, S. W. Nam, H. Park, I. H. Park, N. H. Park, E. S. Seo, R. Sina, J. Wu, J. Yang , et al. (2 additional authors not shown)

Abstract: The CREAM calorimeter, designed to measure the spectra of cosmic-ray nuclei from under 1 TeV to 1000 TeV, is a 20 radiation length (X0) deep sampling calorimeter. The calorimeter is comprised of 20 layers of tungsten interleaved with 20 layers of scintillating fiber ribbons, and is preceded by a pair of graphite interaction targets providing about 0.42 proton interaction lengths (λint). The calori… ▽ More The CREAM calorimeter, designed to measure the spectra of cosmic-ray nuclei from under 1 TeV to 1000 TeV, is a 20 radiation length (X0) deep sampling calorimeter. The calorimeter is comprised of 20 layers of tungsten interleaved with 20 layers of scintillating fiber ribbons, and is preceded by a pair of graphite interaction targets providing about 0.42 proton interaction lengths (λint). The calorimeter was placed in one of CERN's SPS accelerator beams for calibration and testing. Beams of 150 GeV electrons were used for calibration, and a variety of electron, proton, and nuclear fragment beams were used to test the simulation model of the detector. In this paper we discuss the performance of the calorimeter in the electron beam and compare electron beam data with simulation results. △ Less

Submitted 24 March, 2010; originally announced March 2010.

Comments: 29th International Cosmic Ray Conference Pune (2005) 8, 371-374