International Workshop on Hands-Free Speech Communication (HSC2001) April 9-11, 2001 Kyoto, Japan

Hierarchical Feature Compensation and Its Application to Hands-Free ASR

Hui Jiang, Frank Soong, and Chin-Hui Lee

Dialogue Systems Research, Multimedia Communication Research Lab, Bell Labs, Lucent Technologies, Murray Hill, NJ, USA

In this paper, we investigate how to improve the robustness of hands-free speech recognition when only a single or a few test utterances are available for compensation. A new hierarchical tree-based feature transformation is proposed to enhance the conventional stochastic matching in the cepstral feature space. The tree-based hierarchical transformation is estimated from test utterances based on three different criteria: i) maximum likelihood (ML) using the current test utterance; ii) sequential maximum a posterior (MAP) using the current and previous utterances, and iii) structure MAP (SMAP) using current test data. Recognition results obtained using a hands-free database show the proposed feature compensation is robust and significant performance improvement has been observed.

Full Paper

Bibliographic reference.  Jiang, Hui / Soong, Frank / Lee, Chin-Hui (2001): "Hierarchical feature compensation and its application to hands-free ASR", In HSC2001, 111-114.