Initial Implementation of CC generation via HF models

[CosmoWorker]

Below showcases the approach in a mermaid diagram of what I added. From extraction using ffmpeg and using a audio classification model like ast-finetuned dataset model from hugging face. Utilised Opencv visual motion for current behaviour which can be improved.

Here is the sample demo implementation using the models mentioned.

mergedcc.webm

• A deepface model for visual validation is the next step in this process for a better quality and it would describe the emotion of the person as well.
• Speaker diarization can be looked over to segregate segments for later visual score when generating confidedence scores. More such details would be towards proposal