Drop fastai_audio as a dependency for model training

[BrunoGrandePhD]

Currently, model training (i.e. 2_FastAI_StarterScript.ipynb) depends on fastai_audio, which in turn depends on fastai v1. This limits us to old package versions across the board (including pytorch). This complicates the setup of a working environment and forces us to use outdated libraries.

The goal of this issue is to drop fastai_audio as a dependency so we can start migrating to more recent versions of fastai (v2).

[BrunoGrandePhD]

After looking into this, I think this should be part of #214. I don't think it's worthwhile to drop fastai_audio while keeping fastai v1.