7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Incremental On-Line Feature Space MLLR Adaptation for Telephony Speech Recognition

Yongxin Li, Hakan Erdogan, Yuqing Gao, Etienne Marcheret

IBM T.J. Watson Research Center, USA

In this paper, we present a method for incremental on-line adaptation based on feature space Maximum Likelihood Linear Regression (FMLLR) for telephony speech recognition applications. We explain how to incorporate a feature space MLLR transform into a stack decoder and perform on-line adaptation. The issues discussed are as follows: collecting adaptation data on-line and in real time; mapping adaptation data from previous feature space to the present feature space; and smoothing adaptation statistics with initial statistics based on original acoustical model to achieve stability. Testing results on various systems demonstrate that on-line incremental FMLLR adaptation could be an effective and stable method when the adaptation statistics are mapped and smoothed.

Full Paper

Bibliographic reference.  Li, Yongxin / Erdogan, Hakan / Gao, Yuqing / Marcheret, Etienne (2002): "Incremental on-line feature space MLLR adaptation for telephony speech recognition", In ICSLP-2002, 1417-1420.