Thank you! I rebuilt the utts.mlf file and it worked. And I found the two voices are not completely identical–sometimes different units get selected, but maybe of the same level of naturalness.
Hi! I tried to train HMMs on a subset of the original data as well. I took a quarter of the original data, and found that the new voice sounded identical to the original one. Is it because the amount of training data is not small enough for us to hear a difference? Also, I noticed that in the do_alignment script, HVite also uses the file train.scp. Should I change that to “train_subset.scp” as well (the instruction only mentioned HCompV and HERest)?
I also encountered an error when I tried to run the modified do_alignment script at first:
“ERROR [+6510] LOpen: Unable to open label file /mnt/courses.homes/s1551548/Documents/ss_ori/mfcc/alicia_a001.lab”
And I found that this only happened for my own script. After I deleted the mfcc entries of my own recordings, the scp file and the do_alignment script worked. Could you suggest where might the problem be?
Thanks!
Some forums are only available if you are logged in. Searching will only return results from those forums if you log in.
Copyright © 2024 · Balance Child Theme on Genesis Framework · WordPress · Log in