The audio pipeline for any speech process is a simple serial chain, but what you wish to use and what opensource can be extremely varied.

I feel there needs to be an opensource common set of containers linked by a network protocol that is a simple route and queue.
That can dump audio to an ALSA file based source or file, but using standard audio frameworks than proprietary solutions with code needs.

I think websockets due to it 2 packets types is optimum but that is just my opinion of binary/text to carry audio and a file of context.

Each container contains a recieving container and outgoing client that can carry stream or file and queue when the forward container is busy.
The input is a server to support parallel processes that might need to combine to share resources with a client that can route to several servers in a routing queue to allow sclaing if busy.

This is so any opensource speech processing software can be intergrated with no code requirements or at least minimal requirements so a linux voice system can be inclusive of choice of all frameworks than any singular.

I think Linux Voice containers likely docker or common container based system, but I have no preference.

I have posted here due to the strong higher level educational links that should be free of the self interest product based 'opensource' often displays in terms of self interest.

LVC was just my own thoughts on how this could be implemented in the most simplest way of the same common denominator code base.

I actually can not provide this due to my MS amnesia where I could likely create but my memory would fail for support.
It would be really beneficial for a system for any voice process to be easily chained into a system of choice. That is a simple linking of various speech processing software that because of many dependencies partioning into containers with a simple audio protocol / file exchange should mean inclusion of all.

I have mentioned ALSA file based sources/sinks but any common OS supported sound framework than proprietary and same with websockets as any that has common available libs that is cross platform, which is prob easier than virtual sources that might need platform specifics.

A file of the last context of the previous container is all that is needed for any system some form of json that has no fixed protocol apart from the ability to append context at each stage.