Add support for AMD ROCm hardware-accelerated Whisper

[Hedda]

Add out-of-the-box support for AMD ROCm hardware-accelerated Whisper to this Linux Voice Assistent project.

Similar to example this AMD ROCm build of Wyoming Faster Whisper with a precompiled CTranslate2 ROCm fork that has it baked in, targeting gfx900–gfx1151 including Strix Halo (AMD Ryzen™ AI Max+ 395 with Radeon 8060S GPU).

• https://rocm.blogs.amd.com/artificial-intelligence/ctranslate2/README.html

• https://github.com/arlo-phoenix/CTranslate2-rocm

• ROCM support for AMD GPUs m-bain/whisperX#566

See example Wyoming Faster Whisper on AMD ROCm GPUs with prebuilt CTranslate2:

https://hub.docker.com/r/pigeekcom/wyoming-faster-whisper-rocm

[lowlyocean]

I don't think you need any changes from this project, because this project doesn't do any of the Whisper processing in its codebase.

Have you already tried deploying that Whisper docker container on the same machine, and then adding a Wyoming integration in Home Assistant that connects to it? Then in your "Full local assistant" that gets created by adding Satellite to HA, you choose that Wyoming integration