Sentence examples for adaptation matrix from inspiring English sources

In this paper, an adaptation matrix in an NMF framework is introduced to adapt the source dictionary to the target dictionary.

This adaptation matrix is estimated using only a small parallel speech corpus.

In this method, the adaptation matrix is estimated from the fixed source dictionary and source activity matrix, and target speaker information is extracted using the adaptation matrix in this procedure.

As shown in Fig. 5, the proposed model performed best with any type of the adaptation matrix.

In other words, the adaptation matrix is independent of the phoneme, and it is the conversion matrix from the source to the target speaker.

In addition, to facilitate analysis, we assume that all elements of the adaptation matrix C are equal to (frac {1}{N}).

The nonadiagonal adaptation matrices performed the worse for all gender types.

As shown in Fig. 7, we achieved better results as the number of diagonals in the adaptation matrices increased and the best when using the heptadiagonal adaptation matrices, except for F2F conversion.

Interestingly, the proposed model achieved a high MDIR even when the diagonal adaptation matrices were used, unlike the ARBM and linear approaches.

Table 3 compares the performance for each speaker pair with different criteria of the MDIR and MCD, using the SATBM with heptadiagonal adaptation matrices.

Fig. 7 Comparison of VC performance of the proposed model with changing the type of adaptation matrices for each gender type.