Initial changes required to get this running on OSX 13.5 / Docker Engine 24.X

[NickJLange]

I am enthusiastic about the concept of an open Alex/Siri replacement - thank you for blazing the trail here. I did a clone/build on OSX 13.5 with Docker Engine 24.0.5 with the following fixes to get it to boot. Currently, the WIS server on CPU is only returning "You" - so need to do some additional debugging to figure out where the break down is. I'll raise a separate Issue for that.

[kristiankielhofner]

Can you give some performance stats? The output from warm_models on WIS startup is fine.

In addition to working through your other issue with this (#114) my concern is that support for Mac and Docker Desktop will present additional maintenance and support challenges for a configuration that fundamentally doesn't offer a good user experience.

I'd be more interested in native support on Mac. Most of our performance magic for Whisper comes from ctranslate2 and it has an Apple Accelerate backend. Although I've never tested it I imagine running natively on CPU with or without Apple Accelerate support would provide significantly higher performance and eliminate a lot of the networking, etc issues.

[NickJLange]

I agree with you re: Docker - the point of this tree was to have a benchmark/north star to then attempt to get native M2 CPU/GPU support going.

I actually have another tree where I stripped out Docker and tried to wade into things. For the stripped-out-tree, I can get the services up, but am getting errors on model version mismatches, which indicates a library issue that I ran out of time to debug. If interested, I can post up a separate draft PR - that might speed up some of the triage?

For now, I think the next step is to focus on getting the control version even with subpar perf.

[kristiankielhofner]

I have an older Macbook Pro M1 I could do some development and testing with. I created a new branch where we can track these changes.

I think that native support will not only offer higher performance and easier onboarding but more importantly should really help with the WebRTC issues you are experiencing with #114.