In this paper, we propose a new speaker adaptation algorithm that employs vocal tract length normalization (VTLN_) ith generation-dependent acoustic models, and prove its validity with various genration subjects including small children and aged people.
展开▼