TaskBridge worker for transcribing media files
411
This docker image contains a TaskBridge worker for handling transcribe tasks, see https://github.com/hilderonny/taskworker-transcribe.
docker run --gpus=all --name taskworker-transcribe_ROG_LARGE_V2 hilderonny2024/taskworker-transcribe:v1.1.0 --taskbridgeurl http://192.168.0.2:42000/ --worker ROG_LARGE_V2 --device cuda --model large-v2
The order of the parameters is important!
| Parameter | Description |
|---|---|
--name | Name the docker container should get while running |
--taskbridgeurl | The URL of the TaskBridge to use. When the worker is running on the same machine as the TaskBridge, make sure to use the official IP of the host and not 127.0.0.1! |
--worker | Name of the worker to be shown to the TaskBridge |
--device | cuda for using the GPU, cpu otherwise |
--model | Whisper model to use. Can be one of tiny, base, small, medium, large, large-v2 (recommended), large-v3, distil-large-v3 |
On the first run the container needs an internet connection to download the model (up to 5 GB depending on the model). While downloading the container has no output. This can take up some minutes depending on the model but is normal.
You can watch the progress by opening a terminal in the docker container and running
ls -la /app/models/faster-whisper
| Model | Size | GPU RAM requirement | Remarks |
|---|---|---|---|
| tiny | 72 MB | 400 MB | Not useful for languages other than english |
| base | 139 MB | 400 MB | |
| small | 462 MB | 900 MB | |
| medium | 1.4 GB | 2.8 GB | |
| large-v2 | 2.9 GB | 4.2 GB | Best quality for multiple languages |
| large-v3 | 2.9 GB | 4.9 GB | Inaccurate |
| distil-large-v3 | 1.4 GB | 1.8 GB | Currently only supports english |
Content type
Image
Digest
sha256:58a2e07ec…
Size
8.2 GB
Last updated
4 months ago
docker pull hilderonny2024/taskworker-transcribe:large-v2