Improper TDNN training

[Dhanvi Karanam]

Hello,

I am trying to finetune my already trained TDNN. After plotting the log-probabilities while training, I observed a sudden rise in log-probability after the 491st iteration (the plot is in the attachment). After reading the training logs, I have observed that the --read-cache and --write-cache parameters which are used to specify the locations of the cache files for reading and writing during the training process are different from the 492nd iteration (the locations are highlighted in the 2nd and 3rd images).

I wanted to understand why such is happening. Also, how do I resolve this? 

[School at Home]

Hi
I am part of this group.

We are using Linux , an asterisk with Vosk.
Kindly add me to the vosk support group.
Or can I ask here about Vosk related issues.

thanks

> --
> Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
> ---
> You received this message because you are subscribed to the Google Groups "kaldi-help" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/4f44d0f0-53f0-4c5b-8880-60712d030bd8n%40googlegroups.com.