Setting of the Values of the Parameters, H and M We have ever experimented to measure average cepstral distances (ACD) for different combinations of H (probability threshold) values and M (the number of mixture components in a segmental GMM) values. The measured values are listed in Table A. From Table A, it may be seen that the values of ACD under different combinations of H and M are very close, i.e. not sensitive to H and M. For example, the ACD ratio between (H=0.2, M=8) and (H=0.2, M=16) is 0.5220 / 0.5174 = 1.0089.

Table A   ACD for combinations of H and M
 H value M (no. mix.) 0.1 0.2 0.3 0.4 0.5 8 0.5218 0.522 0.5217 0.5216 0.5215 12 0.5196 0.5196 0.5194 0.5195 0.5195 16 0.5176 0.5174 0.5182 0.5182 0.518 20 0.5203 0.5202 0.52 0.52 0.5196

Setting of the Values of the Parameters, K and α
Since the values of ACD are not consistent with the signal quality of the converted voice, we originally tune the vales of K and α by listening to the converted voices. Now, we know that the measured values of variance ratio (VR) are almost consistent with the converted-voice quality. Therefore, we experiment to decide the values of K and α according to the measured VR values. The measured VR for different K values are listed in Table B whereas the measured VR for different α values are listed in Table C.

From Table B, it is seen that the average VR is not sensitive to K (beam width for dynamic programming) value within the range, 16 to 32, but reach the maximum when K=24. Therefore, we decide to use K=24.

From Table C, it is seen that the average VR is not sensitive to α (weighting factor) value within the range, 0.25 to 6. Thus, we cannot but to suspect whether VR cannot distinguish the converted-voice quality when VR values are close. Also, we cannot distinguish the converted-voice qualities when α is varied from 0.5 to 2 by perception. Therefore, we select to use α=1.5.

Table B  VR measured for different K values (under α=1.5)
 K value 12 16 20 24 28 32 36 MA=>MB 0.6228 0.6239 0.6239 0.6248 0.6245 0.6239 0.6233 MA=>FA 0.5783 0.5798 0.5795 0.5793 0.5791 0.5795 0.5789 Average 0.6006 0.6019 0.6017 0.6021 0.6018 0.6017 0.6011

Table C  VR measured for different α values (under K=28)
 α value 0.25 0.5 1 1.5 2 4 6 MA=>MB 0.6239 0.6241 0.6241 0.6245 0.6246 0.6247 0.6245 MA=>FA 0.5794 0.5796 0.5791 0.5792 0.5789 0.5788 0.5793 Average 0.6017 0.6019 0.6016 0.6019 0.6018 0.6018 0.6019