Spleeter: Model Repeat Download

[pHequals7]

After I ran the demo script using the audio example file, when i tried it with my own music file, it again downloads the previously downloaded model which I feel makes it very slow and inefficient

| OS - Windows
| Installation type - git clone and conda create env
| RAM available - 8GB
| Hardware spec GPU - 1050 CPU - i5 8th Gen

[sjernigan]

If you are using the docker image, read the paragraph starting with 'For using your own audio file you will need to create contain' on the Usage page

[wistephens]

I'm have this need as well. Whenever I attempt to use the 5Stem split I get 5 wav files with basically the same content. The 2 and 4 stem models work very well. I lost network connection when the 5stem model was downloading and I cannot figure out how to force a re-download of the model.

| OS - MAC
| Installation type - git clone and conda create env
| RAM available - 8GB
| Hardware spec CPU - 2.8 GHz Intel Core i7

[wistephens]

I fixed my issue:

• download the 5stem.tar.gz file
• replace contents of spleeter/pretrained_models/5stem with contents of tar.gz

[pHequals7]

I tried the third time using custom audio - 2stems and this time it worked without downloading.