Vosk: Only possible for hassio users?

Configuration
,
1

Hi,

I am trying to use Vosk, but there seems to be “only” a hass.io addon.

I tried building the Dockerfile from here, but since there is no CMD it doesn’t run, I guess it is missing some hass.io magic.

Should I run rhasspy instead?

Thanks!

2

What’s a “Vosk”? A link would help.

3

There is a link to the hass.io addon in the post :slight_smile: It is a speech recognition toolkit.

Here is a link to the underlying vosk-project.

My issue is that I can’t get it to work because it’s a hass.io addon, but I am not using hass.io.

4

The link you posted to the Vosk addon does not work.

there is a link to it here. if It helps anyone.

2 Likes
5

Thanks, Arh!

Weird, the link works on my end…

6

I couldn’t find another solution so I created my own container:

2 Likes
7

Thank you! It works fine for me on my Intel N100.

In my case I use Spanish, do you know how I can configure it to download the large version for Spanish? By default, using “PRELOAD_LANGUAGE=es” downloads the small version, regards.

8

I can’t test this as I am out for work for the next few days, but you should be able to just dump any model in the dir of your language.

So in your case inside your model dir there should be a dir called es and inside that folder you should put the model you want to use. Then restart the container.

Let me know if that works :slight_smile:

1 Like
9

Yes, that’s what I did and it worked, now it understands me perfectly and goes quite fast.
Thank you very much for your contribution :slight_smile:

10

My pleasure :slight_smile:
Would you mind sharing what hardware you are running the large model on? Just in case other people consider using the large dataset with a comparable hardware.

1 Like
11

Sure:
Firebat T8 Pro Plus minipc
Processor: Intel n100
Ram: 8GB
Storage: SSD 256GB

Nowadays there are better ones, what I liked about this one was its low consumption (less than 5W) very similar to Raspberry but with an architecture that does not give so many headaches and better performance.

1 Like
12

hi, i want to use vosk in italian due to the fact that whisper is very poor in my native language.

Can you describe me the steps you did to enhance it?

I installed it with this settings but definitely there’s something wrong:

image
image1077×605 30.3 KB

thanks!

13

See this reply :slight_smile:

14

I checked my /share/ folder and I don’t have any “vosk” folder.

Should i create it and put the extracted files from vosk-model-it-0.22.zip?

15

What does your docker-compose.yaml look like?

16

I realized there were a few things that could be improved (and a few things that were wrong, so I updated the repo). The readme now contains a section on how to choose your model.

Let me know what I can improve:

When starting the container it’ll tell you which model it loaded.

@jonatanpradasn
I checked the advice I gave you and it was probably wrong, I am surprised it works for you. Giving you a ping in case you want to double check.

17

I’ll test tonight if I have time (and if I’m able to do it maybe I’ll create a unraid template to spread the word :slight_smile: )

I keep you posted

18

I tried to use your local compose in my unraid environment without success :sleepy:

19

Fantastic, works much better than faster-whisper, no problems in the installation process, except that it doesn’t use the template you downloaded and put in “dirs/models/it” but uses the “small” template of the language template put in docker-compose.