8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Feature Compensation Technique for Robust Speech Recognition in Noisy Environments

Young Joon Kim (1), Hyun Woo Kim (2), Woohyung Lim (1), Nam Soo Kim (1)

(1) Seoul National University, Korea
(2) Electronics and Telecommunications Research Institute, Korea

In this paper, we analyze the problems of the existing interacting multiple model (IMM) and spectral subtraction (SS) approaches and propose a new approach to overcome the problems of these algorithms. Our approach combines the IMM and SS techniques based on a soft decision for speech presence. Results reported on AURORA2 database show that proposed approach shows 14.26% of average relative improvement compared to the IMM algorithm in the speech recognition experiments.

Full Paper

Bibliographic reference.  Kim, Young Joon / Kim, Hyun Woo / Lim, Woohyung / Kim, Nam Soo (2003): "Feature compensation technique for robust speech recognition in noisy environments", In EUROSPEECH-2003, 357-360.