Abstract
A new speaker adaptation technique for the hidden Markov model (HMM) based on the maximum model distance (Kwong,: 1998; He, 1999 approach is presented. Experimental results have shown that this technique provides good performance even with a small amount of adaptation data. When these results are compared with those from the Baum-Welch approach and the stochastic matching approaches (Sank, 1996), it is found that the presented approach provides the best performance. © IEE 1999.
Original language | English |
---|---|
Pages (from-to) | 1817-1818 |
Journal | Electronics Letters |
Volume | 35 |
Issue number | 21 |
DOIs | |
Publication status | Published - 14 Oct 1999 |
Externally published | Yes |