Training fails when i use $slotname in a sentence

Hi,
When i add this line to my sentences , traing fails without giving more info to work with. It JUST SAYS TRAINING FAILED :frowning:

[getPrixR]
Quel est le prix de ($arts)

This Also doesn’t work :

[getPrixR]
Quel est le prix de (PIGMENT JAUNE | PIGMENT VERT | PIGMENT ROUGE)

SLOT :

{
“arts”: [
“PIGMENT JAUNE”,
“PIGMENT ROUGE”,
“PIGMENT VERT”,
“PIGMENT ORANGE”,
.
.
.
]
}

Thank you for your time :smiley:

Tout en minuscule, ça donne quoi?

1 Like

Well this works :

[getPrixR]
Quel est le prix de (pigment jaune | pigment rouge …)

But i have a lot of values in my slot , and this still doesn’t work :

Quel est le prix de ($arts)

Thank you :smiley:

Moi j’utilise pour Intent Recognition: OpenFST
seul sans Fuzzy text matching
et Speech Recognition: kaldi

Hi @kookic,

I tried your settings it seems the speech recognition is better :smile:. But there is a latency to transform in text (no latency with pocketsphinx).
Do you have the same problem ?

Yes, but I use kaldi with a special french model


and is good speed, and I use .env, not docker, .venv is more speed

Hi,

Do I just have to copy all these files and subdir in the …/kaldi/model folder ?

I did it, but not sure.
After reboot I got this message :

KaldiDecoder Missing HCLG.fst Graph not found at /profiles/fr/kaldi/model/graph/HCLG.fst. Did you train your profile?

download good model fr =
wget https://github.com/pguyot/zamia-speech/releases/download/20190930/kaldi-generic-fr-tdnn_250-r20190930.tar.xz

replace
.config/rhasspy/profiles/fr/kaldi/model/model
with this folder
kaldi-generic-fr-tdnn_250-r20190930/kaldi-generic-fr-tdnn_250-r20190930/model

if good kaldi is installed in rhasppy!
if not delete folder kaldi and restart

All release =
https://github.com/pguyot/zamia-speech/releases/

Hi,
Thank’s I’ll do it in couple of days.

The slots do not accept all special characters … Some cleaning and formating was needed for my slots to work.
i used php to get my slots list , all i needed was to pass each entry through this function :

function clean($string) {
return (preg_replace(’/[^A-ZA-úa-z0-9\ %]/’, " ", $string)); // Removes special chars except for the ‘%’ and the accents.
}

Thanks @kookic for the Hint.

Hi,

Done and work.
Thks @kookic

@kookic I’ve seen your instructions about how to use the french model.

I thought I could take these and use them for this model: Model German

Unfortunately it is not working. How do you know which files you need to copy to .config/rhasspy/profiles/fr/kaldi/model/model?

Do you know how I can use this model?

I’m still getting the same error as @raqpub

KaldiDecoder Missing HCLG.fst Graph not found at /profiles/fr/kaldi/model/graph/HCLG.fst. Did you train your profile?

The files in model dir are:
cmvn_opts
den.fst
final.mdl
graph << directory
normalization.fst
tree

or reinstall kaldi and replace this files by your model

look:
https://goofy.zamia.org/zamia-speech/asr-models/

By reinstall you mean that I delete the kaldi folder by hand and after that click download in the UI again? Then replace the files again.

yes

https://goofy.zamia.org/zamia-speech/asr-models/

Thanks. den.fst and normalization.fst are missing in these downloaded images and they are not in my model folder at all. Haven’t found them after reinstalling kaldi. Do you know how to get them?

generic de
https://goofy.zamia.org/zamia-speech/asr-models/

You mean they are inside the 1.4 Gb image? Because I’ve downloaded TDNN 250. They are not in this image.

wget this 111mo

https://goofy.zamia.org/zamia-speech/asr-models/kaldi-generic-de-tdnn_250-r20190328.tar.xz