New russian model available for download doesn't seem to work

[Otherend]

Hello,

I'm currently trying the new russian model (cmusphinx-ru-5.2) and the results I got from it are awful.
I tried to compare it with the previous model (zero_ru.cd_cont_4000) and used it with pocketsphinx to spot keyphrases, it works great. With the new model it is not detecting the keyphrases.

I'm running this command for the old model (same command with the new model except that I changed the dict and hmm) :
pocketsphinx_continuous -dict ru.dic -hmm zero_ru.cd_cont_4000 -infile decoder-test_.wav -kws keyphrase -time yes -logfn /dev/null

The file keyphrase contains :
илья /1e-10/
ильф евгений /1e-20/
петров /1e-20/
золотой /1e-20/
телёнок /1e-20/

The wav is mono, 16000 Hz and its transcript is : илья ильф евгений петров золотой телёнок

Am I using the new model correctly? Is there a problem with the new model?

Thank you for your consideration,
Otherend

[Nickolay V. Shmyrev]

I am not sure where you found 16khz file, the original was 8khz. The new model is for 16khz.

[Otherend]

I think I know why it did that, I upsampled from 8kHz to 16kHz, my bad...

[Otherend]

Is there anyway to use 8kHz wav on a 16kHz trained model? If not, is there a russian model that is trained on 8kHz?

[Nickolay V. Shmyrev]

Is there anyway to use 8kHz wav on a 16kHz trained model?

No

If not, is there a russian model that is trained on 8kHz?

zero_ru.cd_cont_4000 is 8khz.