Live-Video-Translation Plugin Beta using OpenAI Whisper

[axelsl 19]

Hi all,

If anyone is interested, I coded a similar plugin but using OpenAI APIs instead of it running it through a hosted docker. You need an OpenAI account, and each API use is chargeable (so far I haven't used more than $5 a month). I have not published it yet because it is not as perfect as I want it to be, but if anyone is interested, I could share it.

@bobo99thanks for the inspiration!

Axel

[mrtj18 4]

43 minutes ago, bobo99 said:

Hmm did you run the docker with the Gpu flag?  

 

onerahmet/openai-whisper-asr-webservice:v1.7.0-gpu

 

And pass through Nvidia visible devices ? 

1. Yes I ran the docker container with the onerahmet/openai-whisper-asr-webservice:latest-gpu flag to pull down the correct container

2. Yes with unraid I used the "--runtime=nvidia" flag within the extra parameters section of the docker container and the nvidia visible devices as well as nvidia capabilities flag as well and a combination of those, and that didn't work either.

I did see this topic in the forum, but I had no idea where to make the necessary changes within my setup, so I called it quits. 

https://github.com/ahmetoner/whisper-asr-webservice/issues/312

[Neminem 1518]

@mrtj18I do this with unRaid.

But my GPU is a ancient GTX 1080, using with Bazarr and Lingarr.

Not sure if this helps you.

[Geeked 5]

Tried it out on a couple videos and it works great for about 30 seconds then the video freezes and never starts back up again.

[bobo99 13]

23 minutes ago, Geeked said:

Tried it out on a couple videos and it works great for about 30 seconds then the video freezes and never starts back up again.

What hardware are you using , how much GPU ram does your card have, and what sized model are you using ? 

[mrtj18 4]

3 hours ago, Neminem said:

@mrtj18I do this with unRaid.

But my GPU is a ancient GTX 1080, using with Bazarr and Lingarr.

Not sure if this helps you.

Thanks for the help, but I unfortunately my issue, is whisper asr does not like 50 series GPUs. Not without modifications, and I'm not sure how to do it. Via the link previously provided in my last post. 

Thanks anyway

[kexus 0]

I just tested this.

To get it working with a 5000 series card:

1. Clone the whisper-asr-webservice git repo or download and unzip the source
2. Modify pyproject.toml as described here (change every cu126 to a cu128)
3. Follow the development steps to build and run the service

After that it seems to work fairly well at first, but playback always stops after 4 minutes. If I start a new video it'll work again for another 4 minutes, so it doesn't seem to be a problem with the whisper service stopping. Still, it's a very cool idea and I'd love to see it added as a proper plugin.

[mrtj18 4]

On 12/26/2025 at 5:52 PM, kexus said:

I just tested this.

To get it working with a 5000 series card:

1. Clone the whisper-asr-webservice git repo or download and unzip the source
2. Modify pyproject.toml as described here (change every cu126 to a cu128)
3. Follow the development steps to build and run the service

After that it seems to work fairly well at first, but playback always stops after 4 minutes. If I start a new video it'll work again for another 4 minutes, so it doesn't seem to be a problem with the whisper service stopping. Still, it's a very cool idea and I'd love to see it added as a proper plugin.

Very nice, I will keep an eye on this thread to see if the 4 min bug is resolved. Thanks for the steps 

[bobo99 13]

If it isn't working well, is need to know the things outlined in the first post. 

 

Hardware , model size,  etc and the logs.