[Bug] Windows - separate is getting stuck after about a minute into processing. Not hardware constrained.

- [x] I didn't find a similar issue already open.
- [x] I read the documentation (README AND Wiki)
- [x] I have installed FFMpeg
- [x] My problem is related to Spleeter only, not a derivative product (such as Webapplication, or GUI provided by others)

## Description

`spleeter separate` runs, starts a bunch of python processes, starts ffmpeg, consumes a few GB of ram, starts writing the output files correctly, then just stops all activity forever. If I play accompaniment.wav, it is valid, it just cuts off at about 50seconds and never continues. The app produces no messages in verbose mode beyond the lines I pasted below.

The system has tons of RAM/CPU/Disk space available.

I've tried flac and wav, and tried limiting size with `-d` to no avail.

## Step to reproduce

1. Installed using virtualenv w/ python 3.8, pip install spleeter.
2. Run as `spleeter separate '.\ARIatHOME - Let Me In.wav' -p spleeter:2stems -o $PWD --verbose
3. Got no error, but output incomplete and app runs forever.

## Output

```bash
(venv) PS C:\Users\haze\Downloads\yt> spleeter separate -d 600 '.\ARIatHOME - Let Me In.wav' -p spleeter:2stems -o $PWD --verbose
INFO:tensorflow:Apply unet for vocals_spectrogram
WARNING:tensorflow:From C:\Users\haze\Downloads\yt\venv\lib\site-packages\tensorflow\python\keras\layers\normalization.py:534: _colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Colocations handled automatically by placer.
INFO:tensorflow:Apply unet for accompaniment_spectrogram
INFO:tensorflow:Restoring parameters from pretrained_models\2stems\model
```

## Environment

| | |
| ----------------- | ------------------------------- |
| OS | Windows |
| Installation type | pip |
| RAM available | 19GB of 32GB |
| Hardware spec | RTX3070Ti/ Intel i5-12600K / 32GB RAM |