Add speech-dispatcher as /v1/audio/speech endpoint (as tts)

[MrRulf]

Speech-dispatcher is the service in most linux (client) systems that usually takes care of transforming text to speech and is nowadays also often setup with ai models, piper, kokoro...
Having another one that does the same running in openARC therefor would be unnecessary and only use up resources that are restricted anyway (even more on client devices that don't have massive pools like servers). So I think it would make sense to have the option for openARC to just use the speech-dispatcher instead and be a proxy for it.
This way you keep a single instance for tts running but also have it available at the same openAI API as the rest of the AI features.
Servers of course can keep using models directly in openARC, as server mostly doesn't even have speech-dispatcher (what for really).
Speech dispatcher would just be an optional dependency, setting it without having it installed would just return an error and be fine.

[SearchSavior]

Hey thanks for the issue

Are you saying there is a performance impact in some cases?

[MrRulf]

Yesn't.

Not directly performance.

Speech dispatcher is a system level job/api, not exposed via http, but it's basically for tts. It's also required for tts in quite a few apps because it's the default place, so to say i have tts in openARC already i dont need this is not really an option.
So if you need tts for firefox for example you need speech dispatcher. If you also have openARC running with a tts for the endpoint then you have 2 tts engines running. Or at least saved the files in two places.
Instead i say openARC could just access speech dispatcher and act as a bridge between speech dispatcher and the web endpoint. This way you dont need tts running or setup twice.
Of course only as an option that can be configured, use speech dispatcher or do your own tts in openARC.

Speech dispatcher is installed or recommended on most client/desktop linux systems, basically anything with a desktop environment or a window manager, so this affects quite a few people that want to run openARC directly on their machine, where ram and battery are more limited than on a server.
On a server however speech dispatcher usually is not installed so to not annoy those people and to keep the setup as easy as it is - the option to run a tts in openARC shpuld definitly stay.

So, performance impact - technically having both running means in total more ram is occupied, maybe they even use both a little cpu that then isnt free for something else. Best case it just uses up more storage cuz u have models and stuff installed twice.