Abstract

In this paper, the problem of fast model adaptation and complexity selection for nonnative speaker is investigated. The key challenge lies in reliable complexity selection based on a small amount of adaptation data. A novel technique of combining MDL with pseudo likelihood-based state-tying is proposed to enable model complexity selection from using as little as three adaptation speech sentences. In MDL/PL, MDL is performed on nodes with sufficient adaptation data, and pseudolikelihood based state tying is performed on nodes with insufficient adaptation data. Experiments were performed on WSJ data of six nonnative speakers. The combined model adaptation and complexity selection method led to consistent and significant improvement on recognition accuracy over MLLR, with an average error reduction of 10% when a varying number of adaptation speech sentences were taken from each speaker.

‚Äč