When using the load-dataset CLI command, additional non-audio files are being added (like .DS_Store).
The script should check to make sure it is only adding audio files to the dataset and skip non-audio files.

We should make a Github workflow that auto-deploys the demo image to Heroku.

Something like this: https://github.com/marketplace/actions/deploy-to-heroku#deploy-with-docker

Users should have the ability to compare features across different audio samples.

This would most likely be a separate view, like :/compare?sample_id1=333&sample_id2=444&feature_type=spectral

The “Deltas” features should eventually be expanded into something like a “Feature Manipulation” or “Feature Augmentation” filter, which would allow users to perform different types of transformations or filters across all the files in a dataset. A sandbox would be helpful for users to experiment with adding different manipulation types prior to applying the changes across the dataset.

We should think about expanding the annotations section in the Sample Review grid view.

For now the text property is a free-text field, although it is a JSON field within the DB.

Semantic text representation could be added as a way to help users create more robust annotations as mentioned by Ishibashi et. al in https://dl.acm.org/doi/pdf/10.1145/3377325.3377483.

All url links on feature cards should open in a new tab, instead of navigating away from the current page.

Due to performance concerns, the backend WSGI server should be migrated to ASGI.

There are a couple of decent framework options:

• FastAPI - https://fastapi.tiangolo.com/
• Quart - https://pgjones.gitlab.io/quart/ (this is supposed to be a drop-in Flask replacement)

We would probably want to use uvicorn or hypercorn as well.

There is a ton of duplicated code related to feature extraction and image generation.
It is all currently in src/public/views.py and should be moved into it's own class or module somewhere else.

There should be more documentation (and maybe testing) of the load-dataset command.
This would include docstrings/comments/README updates.

There's an issue with how the MFCC delta's graph gets generated, especially with how the higher-order labels get covered by the above graph. See screenshot below: