Upgrade model training to fastai v2

[BrunoGrandePhD]

The goal of this PR is to figure out how to upgrade the model training code in 2_FastAI_StarterScript.ipynb to use fastai v2 (without fastaudio) and attain equivalent performance. After multiple attempts, I settled on the approach of writing two notebooks in parallel:

1. fastai-migration-notes-v1.ipynb: This is mostly code copied from 2_FastAI_StarterScript.ipynb, but with additional exploration on my part (e.g. to confirm certain config values). I also output the data in this notebook so I can compare and contrast with the equivalent code in the second notebook.
2. fastai-migration-notes-v2.ipynb: This notebook demonstrates model training using fastai v2 without fastaudio. I was able to get an accuracy on the validation set (20%) up to 94.9%, which is slightly higher than the 93.7% achieved in the original 2_FastAI_StarterScript.ipynb notebook. While I don't know how much we can rely on these performance gains, my goal was to achieve similar (or greater) performance to make sure I hadn't introduced a regression. I believe this goal was met.

Fixes #214

[pastorep]

Super nice work to get us moving away from deprecated fastaudio package and toward standard torchaudio package. Especially cool to see your experiments proving viability.

I wonder if it makes sense to stick the notebooks in a notes/ folder, to accentuate that no future dev will run these notebooks. I ack that they're already titled -notes.ipynb, but still could be confusing since they're in same directory as our production code.

[pastorep]

Although it does seem that fastai-migration-notes-v2.ipynb may be a reasonable plug-in replacement for 2_FastAI_StarterScript.ipynb. Is that your intention?

[BrunoGrandePhD]

Given your suggestion in #216, I propose that we close this PR and add both notebooks to the new repo in the notebooks/ subfolder with documentation pointing to 2_FastAI_StarterScript.ipynb in this repo. Once that repo is up and running, we can update the documentation in ModelTraining with a link to the new training code.