[TTS] Filter out silent audio files during preprocessing

[rlangman]

What does this PR do ?

First version of preprocessing ignored audio files where start/end of speech could not be detected. During testing I find that all such examples are silent audio files with no speech. And that most public datasets have several such files. This updates the script to remove such files from the final training manifest.

Collection: [TTS]

Changelog

• Update preprocessing script to remove silent/empty audio files from final training manifest.
• Add overwrite_audio so that is script fails partway, you can resume where you left off instead of having to reprocess the entire dataset.
• Fix parsing of overwrite arguments, as trying to parse them as a bool results in True regardless of what value you provide in the CLI.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

[redoctopus]

LGTM, thanks for the fixes