Speech to text / text to speech preview

[w4ffl35]

I uploaded a preview of the current text to speech / speech to text features, take a look here:

https://www.youtube.com/watch?v=tiA_0UIN0ck

[w4ffl35]

Be sure to use speech to text via headphones for now or else the microphone will pick up on the speaker input and the bot will have a non-stop misinterpreted conversation with itself

[w4ffl35]

This PR improves speech capture 78fc018

[w4ffl35]

You can take a look an another chat preview here. This time between myself and an evil chatbot. This is after decreasing speech generation delay and improving voice capture

https://www.youtube.com/watch?v=AahiwSY_gN0

[w4ffl35]

Today I released a number of improvements for text to speech and LLMs, this includes the ability to interrupt the LLM and text-to-speech models (as well as the stream and request queue) via an interrupt button in the request panel. More importantly this interrupt feature will be triggered when the user starts to speak into their microphone.

This gives the user greater control over text generation results by canceling before the LLM finishes, and makes real-time voice conversations with the AI feel more natural.

You can see a preview of this here: https://www.youtube.com/watch?v=m4AJQORHsew