Do typos from reviews

[JamesOwers]

No description provided.

[apmcleod]

The file "reviews.txt" in overleaf contains up-to date status on reviewer comments, plus some comments and discussion (from me) that aren't listed here. The main points remaining are:

• Evaluate tasks 1,3,4 by degradation type. Measure performance split by degradation type for tasks 1, 3, and 4. #127
• Why 5 second excerpts? (Most are longer due to ties, but still, 2 reviewers mentioned this).
• Some soft assertions like "to our judgement" in the new measure_errors.py explanation.
• Decide about mentioning generation in conclusion or not.
• A few things to add to future work (Support excerpts with multiple degradations #32, voice/track degradations, other tasks).

And one point from me:
We may want to change the title (currently "Symbolic music correction using the MIDI degradation toolkit"). I think it is still possible to do. I'm worried that people will skip the paper if they don't work on symbolic data, and now that we have the Degrader class, and the focus is strongly on AMT data aug/noise, we don't want to miss that audience.

Some options:
"Music Correction using The MIDI Degradation Toolkit"
"Error Correction using The MIDI Degradation Toolkit"
"Musical Error Correction using The MIDI Degradation Toolkit"

[JamesOwers]

Things remaining:

• Add figure explaining pianoroll and command format and slim down text
• Add reference "Investigating the Perceptual Validity of Evaluation Metrics for Automatic Piano Music Transcription. Transactions of the International Society for Music Information Retrieval"
• Add columns to confusion matrix figure with results by class for tasks 3 and 4
• Add introduction to ACME dataset