feat: Initial implementation of Gladia transcriptions to BBB 2.7

[lfzawacki]

Still a work in progress, adds few UI/UX elements to support gladia options and a new message used for configuring the transcription backend.

Should be used with this branch of bbb-transcriptions-controller

HTML5 settings in setting.yml:

To enable Gladia support in the HTML5 client:

 audioCaptions:
   enabled: true
   # makes the CC button visible by default
   alwaysVisibile: true
   provider: gladia

For the transcription pad:

 captions:
   enabled: true
   # Hides the old CC button that was used to open the transcription pad.
   # This makes less sense when used together with transcriptions so we hide it
   showButton: false 

The current transcription settings can be configured via userdata parameters on join. Here are the parameters:

userdata-bbb_transcription_partial_utterances -> Boolean: [true, false]
    Enables/disables the use of partial utterance results. These
    are lower quality results that have the advantage of arriving
    faster than the final, and more accurate, results.
    
    The default value is *true*.

userdata-bbb_transcription_min_utterance_length -> Integer: [0,5]
    This value is measured in seconds and denotes the minimum
    size of a partial utterance for it to be considered as valid. Lower
    values will yield quicker and more inacurrate results and vice-versa.
    This can be a good way to balance speed with more accurate results.

    The default value is *3*

userdata-bbb_transcription_provider -> [webspeech, gladia, vosk]
    Sets the transcription provider for this user. 'webspeech' will use
    the old browser based system which is enabled by default on
    most BBB instances. Both 'gladia' and 'vosk' will make BBB use
    the new 'transcription controller' system. 

TODO

• Long results coming from gladia should be split so we don't lose any text

[jqueguiner]

Long results can be handled with endpointing. We are thinking also to add an option to force transcription to maximum Xsec. It should take the min(speech_with_endpointing_trigger, max_Xsec)

[github-actions]

Automated tests Summary

✅ All the CI tests have passed!

[sonarcloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
57 Code Smells

No Coverage information
2.9% Duplication