Some important issues with docker-jitsi-meet

TyomikJan commented 3 years ago

Hello, Alphacep! We have integrated your solution into docker-jitsi-meet. We used a container with the russian model. Of course, your software runs and works, but it has several issues:

If you don’t toggle the microphone button, the recognition displays in json format.
Duplicate and supplementing sentences are written to the file with the recognized text. It would be better to receive the finally recognized sentence.
During a long conference (we checked it for 40 minutes), captions are displayed for the jitsi client, but an error occurs with the kaldi’s container websocket when it comes to write this text to file.
In addition, recognition has a very low quality and we are ready to participate in the training of your model, if you provide the appropriate instructions.

We found something similar in issues #89 and #80, but there are no official solutions. Do you plan to solve these problems and errors we found? We are very interested in these improvements. Best regards, Artem Vartanov, Developer from Granit-Center

nshmyrev commented 3 years ago

We found something similar in issues #89 and #80, but there are no official solutions.

There is a solution in #89

https://github.com/alphacep/vosk-server/issues/89#issuecomment-783713983

Do you plan to solve these problems and errors we found?

Yes

TyomikJan commented 3 years ago

There is a solution in #89

Thanks for your fast response, I will try that solution. I thought that was just your assumption on how to solve this problem.

TyomikJan commented 3 years ago

Hi, @nshmyrev! I compiled the snapshot. It doesn't duplicate strings now, but not all sentences are displayed. Maybe there is a minimum limit of words in a sentence?

wokawoka commented 2 years ago

@TyomikJan did you manage to solve your problem with vosk and jitsi? We are considering to deploy it and connect it to our jitsi instance, but we are unsure whether it's working or not (no duplicate sentences, no skipped sentences). Thank you!

TyomikJan commented 2 years ago

Hi, @wokawoka! Apologies for the delayed response. No, we left this solution. Now we use the latest jitsi version, but without speech-to-text transcription. I haven't followed this repo for a long time, alphacep may have fixed these issues, so try to set it up.

wokawoka commented 2 years ago

@TyomikJan thank you for your reply. We will try to set it up and see the results as soon as we will have some spare time!

alphacep / vosk-server

Some important issues with docker-jitsi-meet #98