Implemented Audio Steaming

[gee842]

Made this change to introduce the potential for more real-time voice applications

Streaming audio in chunks works, but there is some artifacting. Considering this WIP at the moment

[krishna-f22]

@gee842 's repo https://github.com/gee842/dia/tree/streaming-audio-output works after minor fix in layers.py as suggested in #45

But yes, for now it is very slow and not good for streaming.

[SamuraiBarbi]

I hope there's some kind of way to improve the speed and quality of output for streaming. I'd consider streaming a feature that would be a top 3 need for end users and people wanting to implement dia into their projects.

[gee842]

Quantization will help, I'm also going to try it on different hardware to see if I can get something workable

[harmlessman]

Assuming the input consists of two sentences with a total duration of around 10 seconds, how long does it take for the first audio output to start when using your streaming implementation? I understand the latency depends on GPU performance, but I would like to know the generation time and the GPU you used for testing.