What does this PR do?

The integration tests for the data augmentation portion of the code base are broken. I assume at some point code was changed and due to pure coincidence, and the way the tests were written, they still pass. The main function in all the tests is never called, which means results are never generated. Since there are no results in the results directory, the tests run for filename in glob(os.path.join(output_folder, "save", "*.flac")) however this folder will be empty, meaning the tests will all pass because the test function will return after having done nothing.

This PR fixes and improves these tests. I intend to add the integration tests for the other augmentations not currently covered in another PR.

Main Changes

• Fixed yaml imports i.e. time_domain/freq_domain instead of speech_augment
• changed path related code to use os.path.join
• fixed test/augmentation logic where applicable
• removed babble noise test as there is no babble noise class, and this is generally implemented using the AddNoise augmentation which is already covered
• run the main function in the test function so that results are actually generated during testing
• run test function if script is run using python example_.py instead of main function
• updated all expected files to reflect most recent implementation of all augmentations
• changed absolute tolerance on tensor.allclose to 1e-4 (tested on 3 different machines and this seemed to be reasonable in order to account for variability between execution environments) however I invite and encourage reviewers to test locally to further confirm

Fixes #<issue_number>

PR review