Working a something called Continuous Context

[cyberofficial]

Continuous Context is something I'm currently cooking up, i dont know if it'll be ready for 1.2.1

But right now, seems stable enough, but I personally think its not fully ready yet.

The main idea is whisper allows you to have context before feeding it's input. Whisper supports this feature called "initial prompt" where you can provide context before doing a processing.

What I do is have the first inital prompt be set by the user like "This is a live stream of a gamer." Then the first piece of audio is processed, after that next prompt is in the form of: "This audio section is [<user inital>], the previous detected speech was <prev_original> and was <transcribed/translated> to <prev_<transcribed/translated>>."

This method does seem to work, and create more better outputs.

Here is an example:

What is happening that next sentence in line is being filled with missing context to give the next part more better translations and transcription.

I hope i can release this soon, but still needs tweaking.