"Warning: det <= 0. Using Prasanna's magic numbers in 88 of 668 frames"

[raikarsagar]

Hi,
I am getting the following warning during test sample synthesis after training the model with custom data(cmu arctic prompts with indian accent).

CG test_resynth arctic_b0028
Warning: det <= 0. Using Prasanna's magic numbers in 88 of 668 frames
Warning: det <= 0. Using Prasanna's magic numbers in 88 of 668 frames

Any suggestions towards this issue are welcome
Regards
Sagar

[saikrishnarallabandi]

This is just a warning. Should be okay. Check if the data is imbalanced in phones(some phone has too little data)

…

On Thu, Jan 28, 2021, 23:01 Sagar Raikar ***@***.***> wrote: Hi, I am getting the following warning during test sample synthesis after training the model with custom data(cmu arctic prompts with indian accent). CG test_resynth arctic_b0028 Warning: det <= 0. Using Prasanna's magic numbers in 88 of 668 frames Warning: det <= 0. Using Prasanna's magic numbers in 88 of 668 frames Any suggestions towards this issue are welcome Regards Sagar — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#55>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFO4QMM7IX4BWDJOT4R576TS4IXH5ANCNFSM4WYGEDPQ> .

[ddavout]

Check if your lab/*.sl are correct
It could be a problem of environment variables. I had imposed LC_ALL=C in the ehmm/scripts : I am happy to say there is no more Prasanna's magic